Instructions to use GreenBitAI/Qwen-3-235B-A22B-layer-mix-bpw-4.0-mlx with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MLX
How to use GreenBitAI/Qwen-3-235B-A22B-layer-mix-bpw-4.0-mlx with MLX:
# Download the model from the Hub pip install huggingface_hub[hf_xet] huggingface-cli download --local-dir Qwen-3-235B-A22B-layer-mix-bpw-4.0-mlx GreenBitAI/Qwen-3-235B-A22B-layer-mix-bpw-4.0-mlx
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- LM Studio
| { | |
| "measurement": { | |
| "model.layers.0": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.1": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.2": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.3": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.4": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.5": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.6": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.7": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.8": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.9": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.10": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.11": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.12": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.13": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.14": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.15": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.16": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.17": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.18": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.19": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.20": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.21": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.22": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.23": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.24": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.25": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.26": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.27": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.28": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.29": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.30": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.31": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.32": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.33": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.34": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.35": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.36": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.37": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.38": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.39": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.40": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.41": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.42": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.43": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.44": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.45": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.46": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.47": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.48": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.49": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.50": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.51": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.52": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.53": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.54": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.55": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.56": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.57": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.58": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.59": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.60": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.61": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.62": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.63": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.64": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.65": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.66": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.67": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.68": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.69": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.70": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.71": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.72": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.73": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.74": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.75": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.76": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.77": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.78": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.79": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.80": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.81": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.82": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.83": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.84": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.85": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.86": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.87": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.88": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.89": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.90": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.91": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.92": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.93": { | |
| "accuracy": 1e-06, | |
| "total_bits": 10145341440.0, | |
| "q_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "k_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "v_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_gate": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "moe_expert_down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| } | |
| } | |
| } | |