Instructions to use Sangsang/rewind-run1-sft-model with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Sangsang/rewind-run1-sft-model with Transformers:

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

pipe = pipeline("text-generation", model="Sangsang/rewind-run1-sft-model")
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("Sangsang/rewind-run1-sft-model")
model = AutoModelForCausalLM.from_pretrained("Sangsang/rewind-run1-sft-model", 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 Sangsang/rewind-run1-sft-model with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Sangsang/rewind-run1-sft-model"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Sangsang/rewind-run1-sft-model",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Sangsang/rewind-run1-sft-model

SGLang

How to use Sangsang/rewind-run1-sft-model 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 "Sangsang/rewind-run1-sft-model" \
    --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": "Sangsang/rewind-run1-sft-model",
		"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 "Sangsang/rewind-run1-sft-model" \
        --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": "Sangsang/rewind-run1-sft-model",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use Sangsang/rewind-run1-sft-model with Docker Model Runner:
```
docker model run hf.co/Sangsang/rewind-run1-sft-model
```

You need to agree to share your contact information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Rewind run1 SFT model

The completed supervised fine-tuning checkpoint from run1, before GRPO/RL training. This is a full Qwen3-1.7B checkpoint in BF16 safetensors format, with its tokenizer, chat template, model configuration, and generation configuration.

It was trained on the 584 examples in Sangsang/rewind-run1-sft-data. Training used two epochs, global batch size 8, learning rate 2e-6, and eight warmup steps: 146 optimization steps and 1,168 examples seen. The vocabulary is unchanged; <rewind> is a plain-text action marker, not an added special token.

Loading

from transformers import AutoModelForCausalLM, AutoTokenizer

repo = "Sangsang/rewind-run1-sft-model"
tokenizer = AutoTokenizer.from_pretrained(repo, token=True)
model = AutoModelForCausalLM.from_pretrained(repo, token=True, torch_dtype="auto")

This repository is private; authenticate with an account that has access. To execute the learned operation, an inference controller must intercept <rewind>, remove the last four reasoning sections, and resume from the retained prefix. The training instruction allows one rewind per trajectory, before </think>. A normal Transformers generate() call does not perform that context deletion automatically. Deleted tokens still count against the trajectory's generation budget.

training_summary.json records the saved run settings and SFT summary. manifest.json provides file checksums. No RL checkpoint or optimizer state is included.

Downloads last month: -

Safetensors

Model size

2B params

Tensor type

BF16

Model tree for Sangsang/rewind-run1-sft-model

Base model

Qwen/Qwen3-1.7B-Base

Finetuned

Qwen/Qwen3-1.7B

Finetuned

(1258)

this model

Sangsang
/

rewind-run1-sft-model

You need to agree to share your contact information to access this model

Rewind run1 SFT model

Loading

Model tree for Sangsang/rewind-run1-sft-model

Dataset used to train Sangsang/rewind-run1-sft-model