GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
XINLI1997
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
XINLI1997
Model Tree
Qwen/Qwen3.5-4B
Input Modalities
Output Modalities
Supported Functionality
| Base model | Qwen/Qwen3.5-4B |
| Role | Mathematics expert; frozen teacher of the 4B students |
| Training | GRPO from the base model, 320 updates, seed 42 |
| Precision | bfloat16 |
| Chat format | non-thinking (enable_thinking=False) |
| License | Apache-2.0 (same as the base model) |
The full recipe, with the launch scripts for every row of the paper's tables, is in
recipes/qwen3.5/ and docs/recipe.md.
This model was trained and evaluated with the non-thinking chat format. Pass enable_thinking=False to the chat
template. Qwen3.5-4B's chat template enables thinking by default, so this argument is required. The evaluation settings in the paper were temperature 1.0 and top-p 1.0, with up to
16,384 new tokens (8,192 in the appendix).
vLLM (the paper used vLLM 0.18.0):
python
from vllm import LLM, SamplingParamsllm = LLM(model="XINLI1997/DN-MOPD-Qwen3.5-4B-teacher-math", max_model_len=32768)params = SamplingParams(temperature=1.0, top_p=1.0, max_tokens=16384, seed=42)messages = [{"role": "user", "content": "Find the sum of all positive divisors of 36. Put the final answer in \\boxed{}."}]outputs = llm.chat(messages, params, chat_template_kwargs={"enable_thinking": False})print(outputs[0].outputs[0].text)
Transformers (Qwen3.5 needs transformers>=5; the paper's training environment used 5.12.1):
python
import torchfrom transformers import AutoModelForImageTextToText, AutoTokenizertokenizer = AutoTokenizer.from_pretrained("XINLI1997/DN-MOPD-Qwen3.5-4B-teacher-math")model = AutoModelForImageTextToText.from_pretrained("XINLI1997/DN-MOPD-Qwen3.5-4B-teacher-math", dtype=torch.bfloat16, device_map="auto")messages = [{"role": "user", "content": "Write a Python function that returns the n-th Fibonacci number."}]text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True, enable_thinking=False)inputs = tokenizer(text, return_tensors="pt").to(model.device)output = model.generate(**inputs, max_new_tokens=4096, do_sample=True, temperature=1.0, top_p=1.0)print(tokenizer.decode(output[0, inputs["input_ids"].shape[1]:], skip_special_tokens=True))
Paper Table 2 (Qwen3.5-4B; each domain averages its two tasks: AIME25/AIME26, LiveCodeBench v5/v6, IFEval/IFBench):
| Model | Math | Code | IF | Total |
|---|---|---|---|---|
| DN-MOPD-Qwen3.5-4B-teacher-math (Math expert) | 54.2 | 39.7 | 54.0 | 49.3 |
| Initial student (Qwen3.5-4B) | 52.2 | 38.1 | 52.8 | 47.7 |
Scores (%) from the paper; training seed 42; 16,384-token evaluation cap; non-thinking chat template; temperature 1.0, top-p 1.0, generation seed 42. AIME25/AIME26: avg@64. LiveCodeBench v5/v6 (167/175 disjoint problems): avg@6. IFEval/IFBench: strict prompt accuracy, avg@16. Total: mean of the six task scores.
Qwen3_5ForConditionalGeneration, bfloat16), exported from the FSDP training
checkpoint.mtp.*) of the base model. All other tensors have the
base model's names and shapes. MTP-based speculative decoding is therefore not available with this checkpoint.
Ordinary decoding is unaffected: the paper's evaluations used exactly these files.config.json, the tokenizer files and chat_template.jinja are the base model's, unchanged.LICENSE is the base model's Apache-2.0 license.bibtex
@article{li2026dnmopd,title = {Beyond Teacher Assignment: Domain-Normalized Multi-Teacher On-Policy Distillation},author = {Li, Xin and Jiang, Hao and Gao, Xin and Wang, Annan and Xie, Yuchen and Guo, Jinghao and Qu, Xingwei and Zhang, Yichi and Yuen, Chau},journal = {arXiv preprint arXiv:2609.35347},year = {2026},url = {https://arxiv.org/abs/2609.35347}}
This model is a fine-tuned derivative of Qwen/Qwen3.5-4B by the Qwen team, released under the Apache License 2.0.