Instructions to use GreenBitAI/DeepSeek-R1-Distill-Llama-8B-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/DeepSeek-R1-Distill-Llama-8B-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 DeepSeek-R1-Distill-Llama-8B-layer-mix-bpw-4.0-mlx GreenBitAI/DeepSeek-R1-Distill-Llama-8B-layer-mix-bpw-4.0-mlx
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- LM Studio
| { | |
| "measurement": { | |
| "model.layers.0": { | |
| "accuracy": 0.87518310546875, | |
| "total_bits": 889454592.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "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 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.1": { | |
| "accuracy": 0.95147705078125, | |
| "total_bits": 889454592.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "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 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.2": { | |
| "accuracy": 0.980438232421875, | |
| "total_bits": 889454592.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "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 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.3": { | |
| "accuracy": 0.97711181640625, | |
| "total_bits": 889454592.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "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 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.4": { | |
| "accuracy": 0.9727783203125, | |
| "total_bits": 889454592.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "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 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.5": { | |
| "accuracy": 0.964508056640625, | |
| "total_bits": 889454592.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "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 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.6": { | |
| "accuracy": 0.957672119140625, | |
| "total_bits": 889454592.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "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 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.7": { | |
| "accuracy": 0.962005615234375, | |
| "total_bits": 889454592.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "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 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.8": { | |
| "accuracy": 0.96429443359375, | |
| "total_bits": 889454592.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "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 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.9": { | |
| "accuracy": 0.958831787109375, | |
| "total_bits": 889454592.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "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 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.10": { | |
| "accuracy": 0.96038818359375, | |
| "total_bits": 889454592.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "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 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.11": { | |
| "accuracy": 0.959991455078125, | |
| "total_bits": 889454592.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "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 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.12": { | |
| "accuracy": 0.957183837890625, | |
| "total_bits": 889454592.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "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 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.13": { | |
| "accuracy": 0.954437255859375, | |
| "total_bits": 889454592.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "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 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.14": { | |
| "accuracy": 0.959442138671875, | |
| "total_bits": 889454592.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "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 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.15": { | |
| "accuracy": 0.94647216796875, | |
| "total_bits": 889454592.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "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 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.16": { | |
| "accuracy": 0.9613037109375, | |
| "total_bits": 889454592.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "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 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.17": { | |
| "accuracy": 0.9733734130859375, | |
| "total_bits": 889454592.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "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 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.18": { | |
| "accuracy": 0.9786529541015625, | |
| "total_bits": 889454592.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "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 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.19": { | |
| "accuracy": 0.97515869140625, | |
| "total_bits": 889454592.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "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 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.20": { | |
| "accuracy": 0.9787445068359375, | |
| "total_bits": 889454592.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "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 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.21": { | |
| "accuracy": 0.971771240234375, | |
| "total_bits": 889454592.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "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 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.22": { | |
| "accuracy": 0.979736328125, | |
| "total_bits": 889454592.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "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 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.23": { | |
| "accuracy": 0.983734130859375, | |
| "total_bits": 889454592.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "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 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.24": { | |
| "accuracy": 0.977630615234375, | |
| "total_bits": 889454592.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "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 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.25": { | |
| "accuracy": 0.9701690673828125, | |
| "total_bits": 889454592.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "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 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.26": { | |
| "accuracy": 0.9687042236328125, | |
| "total_bits": 889454592.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "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 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.27": { | |
| "accuracy": 0.970367431640625, | |
| "total_bits": 889454592.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "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 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.28": { | |
| "accuracy": 0.9718780517578125, | |
| "total_bits": 889454592.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "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 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.29": { | |
| "accuracy": 0.96551513671875, | |
| "total_bits": 889454592.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "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 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.30": { | |
| "accuracy": 0.959747314453125, | |
| "total_bits": 889454592.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "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 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| }, | |
| "model.layers.31": { | |
| "accuracy": 0.9403076171875, | |
| "total_bits": 889454592.0, | |
| "o_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "down_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "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 | |
| }, | |
| "gate_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| }, | |
| "up_proj": { | |
| "group_size": { | |
| "4": 128 | |
| }, | |
| "bits": [ | |
| 4 | |
| ], | |
| "bits_prop": [ | |
| 1 | |
| ], | |
| "scale_bits": 4, | |
| "scale_groups:": 32 | |
| } | |
| } | |
| } | |
| } |