Parameter	Value
direction_index	20.06
linear_attn.out_proj.max_weight	0.93
linear_attn.out_proj.max_weight_position	37.63
linear_attn.out_proj.min_weight	0.71
linear_attn.out_proj.min_weight_distance	4.94
moe_experts.down_proj.max_weight	1.18
moe_experts.down_proj.max_weight_position	23.68
moe_experts.down_proj.min_weight	0.86
moe_experts.down_proj.min_weight_distance	5.48
shared_expert.down_proj.max_weight	1.33
shared_expert.down_proj.max_weight_position	36.08
shared_expert.down_proj.min_weight	0.70
shared_expert.down_proj.min_weight_distance	17.75
attn.o_proj.max_weight	0.83
attn.o_proj.max_weight_position	35.00
attn.o_proj.min_weight	0.06
attn.o_proj.min_weight_distance	15.22

I suggest using the following sets of sampling parameters depending on the mode and task type:
- Thinking mode for general tasks:
  temperature=1.0, top_p=0.95, top_k=20, min_p=0.0, presence_penalty=1.5, repetition_penalty=1.0
- Thinking mode for precise coding tasks (e.g., WebDev):
  temperature=0.6, top_p=0.95, top_k=20, min_p=0.0, presence_penalty=0.0, repetition_penalty=1.0
- Instruct (or non-thinking) mode for general tasks:
  temperature=0.7, top_p=0.8, top_k=20, min_p=0.0, presence_penalty=1.5, repetition_penalty=1.0
- Instruct (or non-thinking) mode for reasoning tasks:
  temperature=1.0, top_p=1.0, top_k=40, min_p=0.0, presence_penalty=2.0, repetition_penalty=1.0
For supported frameworks, you can adjust the presence_penalty parameter between 0 and 2 to reduce endless repetitions. However, using a higher value may occasionally result in language mixing and a slight decrease in model performance.

This model was converted to MLX format from brayniac/Qwen3.5-35B-A3B-heretic using mlx-vlm version 0.4.1.

TheCluster/Qwen3.5-35B-A3B-Heretic-MLX-9bit

Qwen3.5-35B-A3B Heretic