Qwen3.5-4B-MBT-R

Qwen/Qwen3.5-4B-Base fine-tuned on traces edited by MBT-R: the editor rewrites the whole trace in one call. It accompanies the paper "Distilling Metacognitive Behaviors through Agentic Editing" (anonymous submission).

Model	Training data
Qwen3.5-4B-Unedited-SFT	Unedited teacher traces
Qwen3.5-4B-MBT-R	MBT-R (rewrite)
Qwen3.5-4B-MBT-I	MBT-I (insert)
Qwen3.5-4B-MBT-A	MBT-A (agentic)

Training data: built from the traces in metacognitive-behavioral-tuning/mbt-traces. Code and prompts: GitHub.

Usage

The chat template that ships with the model opens a <think> block, so the generated text is the reasoning, then </think> and the answer. End the question with the instruction used in training:

from vllm import LLM, SamplingParams

llm = LLM("metacognitive-behavioral-tuning/Qwen3.5-4B-MBT-R")
question = "Find the remainder when 2^100 is divided by 7."
prompt = question + "\n\nPlease reason step by step, and put your final answer within \\boxed{}."
params = SamplingParams(temperature=1.0, top_p=0.95, max_tokens=122880)
out = llm.chat([{"role": "user", "content": prompt}], params)
print(out[0].outputs[0].text)

The weights are bf16. Long answers are common, so leave room for tens of thousands of tokens.

Training

Setting	Value
Base model	Qwen/Qwen3.5-4B-Base
Method	Full-parameter supervised fine-tuning, loss on assistant tokens only
Epochs	15
Optimizer	Muon (Megatron implementation), learning rate 1e-5, weight decay 0
Schedule	Constant with 40 warmup steps
Global batch	64 sequences (8 GPUs x 8 accumulation steps), padding-free packing
Maximum length	131,072 tokens
Precision	bf16 autocast with fp32 master weights; released in bf16

License

Apache-2.0, following the base model. The training data is CC BY-SA 4.0; see the dataset card for upstream attribution.

Citation

@misc{mbt2026,
  title  = {Distilling Metacognitive Behaviors through Agentic Editing},
  author = {Anonymous},
  year   = {2026},
  note   = {Under review}
}