Instructions to use Levi980623/Qwen3-4B-Plan-SFT with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Levi980623/Qwen3-4B-Plan-SFT with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="Levi980623/Qwen3-4B-Plan-SFT")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("Levi980623/Qwen3-4B-Plan-SFT")
model = AutoModelForCausalLM.from_pretrained("Levi980623/Qwen3-4B-Plan-SFT", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=256)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use Levi980623/Qwen3-4B-Plan-SFT with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Levi980623/Qwen3-4B-Plan-SFT"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Levi980623/Qwen3-4B-Plan-SFT",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Levi980623/Qwen3-4B-Plan-SFT

SGLang

How to use Levi980623/Qwen3-4B-Plan-SFT with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "Levi980623/Qwen3-4B-Plan-SFT" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Levi980623/Qwen3-4B-Plan-SFT",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "Levi980623/Qwen3-4B-Plan-SFT" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Levi980623/Qwen3-4B-Plan-SFT",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use Levi980623/Qwen3-4B-Plan-SFT with Docker Model Runner:
```
docker model run hf.co/Levi980623/Qwen3-4B-Plan-SFT
```

Qwen3-4B-Plan-SFT

SURE-style Stage 1 strategy-guided SFT of Qwen/Qwen3-4B: given a math problem, the model first writes a concise high-level strategy, then an execution that follows it, then the final answer in \boxed{}. Full-parameter fine-tune (no LoRA), one causal LM, one autoregressive response; no planner/executor bridge, no latent/RL stage.

Output format (Qwen3-native thinking container)

<think>
<strategy>
...high-level plan...
</strategy>
<execution>
...detailed reasoning...
</execution>
</think>

\boxed{...}

Prompt (user turn), applied with add_generation_prompt=True, enable_thinking=True:

Solve the following mathematics problem.
First write a concise high-level plan inside <strategy>...</strategy>.
Then follow that plan in <execution>...</execution> with detailed reasoning.
Finally provide the answer in \boxed{...}.
Do not include the answer or detailed calculations in the strategy.

Problem:
{question}

Decoding used for all evaluations below: temperature 0.6, top_p 0.95, top_k 20, max 8192 new tokens.

from transformers import AutoModelForCausalLM, AutoTokenizer
tok = AutoTokenizer.from_pretrained("Levi980623/Qwen3-4B-Plan-SFT")
model = AutoModelForCausalLM.from_pretrained("Levi980623/Qwen3-4B-Plan-SFT", dtype="auto", device_map="auto")
msgs = [{"role": "user", "content": PROMPT.format(question="What is the sum of the first 100 positive integers?")}]
ids = tok.apply_chat_template(msgs, add_generation_prompt=True, enable_thinking=True, return_tensors="pt").to(model.device)
out = model.generate(ids, max_new_tokens=4096, do_sample=True, temperature=0.6, top_p=0.95, top_k=20)
print(tok.decode(out[0, ids.shape[1]:], skip_special_tokens=False))

Training data

Source: open-thoughts/OpenThoughts3-1.2M @ 61bcf9d4eb38b30295efc2021227a63cc5bb34c8, math domain only, seed 42 streaming sample, deduplicated by normalized question; 9800 train / 200 validation rows.
Execution = the reference write-up after </think> (300–1,500 tokens); answer = its \boxed{}. Full-row lengths for this tokenizer: max 2994 tokens, mean 808.0 (cap 8192, nothing truncated).
Strategies were extracted from (question, reference) by Qwen/Qwen3-235B-A22B-Instruct-2507-FP8 (revision e156cb4efae4, the official FP8 release of the 235B teacher, self-hosted with vLLM, temperature 0), gated by hard rules (no answer leakage, no tags) and a consistency judge; 2219 rows use a single automatic re-extraction after a leak.
Benchmark contamination filter (exact + 8-gram near-duplicate) against MATH-500, AIME24, AIME25, AMC23: all checked.
reference_verification per row (advisory 235B judge, not a gold check): judge_approved 3664, judge_flagged 6336, unverified 0. OpenThoughts3 has no gold answers; no accuracy is claimed.

Training (SURE Appendix D / Table 8 Stage-1 settings)

epochs	lr	warmup	global batch	micro batch	max len	precision	optimizer	sched	wd	clip	distributed
3	4e-05	0.05	32	1	8192	bf16 compute / fp32 master	AdamW	cosine	0.01	1.0	FSDP2, 2× L40S

921 optimizer steps, completion-only token-weighted CE (prompt masked), activation checkpointing, wall-clock ≈ 11.7 h. Final train loss 0.073; validation loss 0.506 (epoch-1 ≈ 0.42: the third epoch over-fits in the usual 3-epoch-SFT way; kept as prescribed).

Evaluation on 200 held-out validation questions (never trained on)

test	n	result
fresh-process reload (`from_pretrained`)	–	PASS; 24/24 sampled tensors changed vs base
smoke generation format success	20	95.0% (boxed answer present 95.0%)
natural generation: strategy / execution / transition / answer parse	100	100.0% / 100.0% / 100.0% / 100.0%
natural generation: matches reference `\boxed{}`	100	24.0% (reference-match, not gold accuracy)
self-plan adherence, rubric v1 (strict, penalises arithmetic slips)	100	16.5% followed; undecidable/judge-error 9.0%
self-plan adherence, rubric v2 (method-naming)	100	55.0% followed
fixed-plan continuation: format / adherence v1 / v2	50	100.0% / 23.9% / 56.0%
valid Plan-A/Plan-B intervention: pairs / method switch success	26	26 valid of 30 (4 not constructible) / 53.8%

Judge for the adherence tests: Qwen/Qwen3-235B-A22B-Instruct-2507-FP8 (non-thinking, temperature 0). Status flag: LOW_PLAN_ADHERENCE — the task's acceptance rule treats adherence below 0.70 as low; rubric v2 gives 55.0%, so this is reported as a finding, not hidden. Caveat: under rubric v1 the judge marked most executions "contradicted" for arithmetic or verification mistakes even when the execution used exactly the strategy's method; rubric v2 asks it to name the method actually used and compare methods only. Both are reported; neither is a ground-truth measurement. Format compliance is near-perfect, answer agreement with the (unverified) references is low — this checkpoint is a Stage-1 initialization for later latent-planning work, not a tuned math solver.

Provenance

Pipeline, tests, dry-runs, decisions and blockers: /PHShome/yl463/sure_stage1_autonomous/ (RESULTS.md, wiki/decisions.md). Card regenerated 2026-10-10T01:29:07-0400 from the evaluation files; weights unchanged since the training export.

Downloads last month: 36

Safetensors

Model size

4B params

Tensor type

BF16

Model tree for Levi980623/Qwen3-4B-Plan-SFT

Base model

Qwen/Qwen3-4B-Base

Finetuned

Qwen/Qwen3-4B

Finetuned

(1168)

this model

Levi980623
/

Qwen3-4B-Plan-SFT