Instructions to use Minkow/TipCoder-RL with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Minkow/TipCoder-RL with Transformers:

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

pipe = pipeline("text-generation", model="Minkow/TipCoder-RL")
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("Minkow/TipCoder-RL")
model = AutoModelForCausalLM.from_pretrained("Minkow/TipCoder-RL", 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 Minkow/TipCoder-RL with vLLM:

Install from pip and serve model

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

Use Docker

docker model run hf.co/Minkow/TipCoder-RL

SGLang

How to use Minkow/TipCoder-RL 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 "Minkow/TipCoder-RL" \
    --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": "Minkow/TipCoder-RL",
		"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 "Minkow/TipCoder-RL" \
        --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": "Minkow/TipCoder-RL",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use Minkow/TipCoder-RL with Docker Model Runner:
```
docker model run hf.co/Minkow/TipCoder-RL
```

TIPCoder-RL

Release candidate for the TIPCoder instruction proposer. This is a proposer checkpoint, not a standalone code generator.

Paper: https://aclanthology.org/2026.findings-emnlp.495/ Code: https://github.com/Minkow/TipCoder

Checkpoint

This export is from actor step 10 of the final RL training run. The proposer was initialized from our SFT model, trained from Qwen3-4B-Instruct-2507. The training lineage was checked against the historical SFT and RL job arguments. The configuration declares Qwen3ForCausalLM with tied embeddings. Historical training-directory names are not architecture identifiers.

The original PyTorch shards repeat large backing storages. This export stores the same 399 named tensors in compact safetensors shards. Every exported tensor was compared byte-for-byte with the original cast to BF16, matching the inference dtype recorded in the experiment's vLLM initialization logs. This is an explicit FP32-to-BF16 conversion, not an identical FP32 archive or integer quantization. The original configuration and tokenizer files are unchanged, including the configuration's BF16 loading hint. The index includes both names for tied weights.

Use the proposer completion template in the code repository. Do not substitute a generic chat prompt without checking the expected training format.

License and scope

TIPCoder contributions follow the MIT license, consistent with AceCoder. Underlying model components retain applicable upstream licensing and notices; this notice does not replace the base model's Apache-2.0 license, which must accompany distribution of the derived checkpoint.

This candidate has undergone BF16 tensor equality checks, not a new benchmark run. No training datasets, optimizer state, logs, tokens or platform credentials are included.

Downloads last month: 74

Safetensors

Model size

4B params

Tensor type

BF16

Model tree for Minkow/TipCoder-RL

Base model

Qwen/Qwen3-4B-Instruct-2507

Finetuned

(2727)

this model