Parameter	Value
direction_index	17.83
attn.o_proj.max_weight	1.05
attn.o_proj.max_weight_position	21.08
attn.o_proj.min_weight	0.80
attn.o_proj.min_weight_distance	19.87
mlp.down_proj.max_weight	1.49
mlp.down_proj.max_weight_position	25.85
mlp.down_proj.min_weight	0.05
mlp.down_proj.min_weight_distance	18.91

Performance

Metric	This model	Original model (a model)
KL divergence	0.0025	0 (by definition)
Refusals	4/100	97/100

Qwen3 4B Thinking 2507 - MiniMax M2.1 Distill

This model was trained on a reasoning dataset of MiniMax M2.1.

🧬 Datasets:
- TeichAI/MiniMax-M2.1-8800x
🏗 Base Model:
- unsloth/Qwen3-4B-Thinking-2507
⚡ Use cases:
- Coding
- Science
- Deep Research
∑ Stats (Dataset)
- Costs: $ 42.94 (USD)
- Total tokens (input + output): 39.2 M

This qwen3 model was trained 2x faster with Unsloth and Huggingface's TRL library.