▀█▀ █▄ ▄█ █ █ █▀▀ █▀▄ █▀▀ ▀█▀ ▀█▀ █▀▀ █ █ ▀ █ █▀█ █▀▀ █▀▄ █▀▀ █ █ █ ▀▀▀ ▀ ▀ ▀ ▀ ▀▀▀ ▀ ▀ ▀▀▀ ▀ ▀▀▀ ▀▀▀

Abliterated/Heretic dphn/Dolphin3.0-Qwen2.5-3b

Check Quants

Refusals (this model): 4/100
Original (dphn/Dolphin3.0-Qwen2.5-3b): 47/100
KL divergence: 0.0804

Parameters
direction_index = per layer
attn.o_proj.max_weight = 0.88
attn.o_proj.max_weight_position = 21.47
attn.o_proj.min_weight = 0.55
attn.o_proj.min_weight_distance = 1.60
mlp.down_proj.max_weight = 1.01
mlp.down_proj.max_weight_position = 22.57
mlp.down_proj.min_weight = 0.68
mlp.down_proj.min_weight_distance = 17.22


Downloads last month
50
Safetensors
Model size
3B params
Tensor type
BF16
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for hereticness/Heretic-Dolphin3.0-Qwen2.5-3b

Base model

Qwen/Qwen2.5-3B
Finetuned
(4)
this model
Quantizations
2 models