Instructions to use zhuyaoyu/CodeV-R1-RL-Qwen-7B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use zhuyaoyu/CodeV-R1-RL-Qwen-7B with Transformers:

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

pipe = pipeline("text-generation", model="zhuyaoyu/CodeV-R1-RL-Qwen-7B")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("zhuyaoyu/CodeV-R1-RL-Qwen-7B")
model = AutoModelForCausalLM.from_pretrained("zhuyaoyu/CodeV-R1-RL-Qwen-7B", 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=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Inference
Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use zhuyaoyu/CodeV-R1-RL-Qwen-7B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "zhuyaoyu/CodeV-R1-RL-Qwen-7B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "zhuyaoyu/CodeV-R1-RL-Qwen-7B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/zhuyaoyu/CodeV-R1-RL-Qwen-7B

SGLang

How to use zhuyaoyu/CodeV-R1-RL-Qwen-7B 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 "zhuyaoyu/CodeV-R1-RL-Qwen-7B" \
    --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": "zhuyaoyu/CodeV-R1-RL-Qwen-7B",
		"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 "zhuyaoyu/CodeV-R1-RL-Qwen-7B" \
        --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": "zhuyaoyu/CodeV-R1-RL-Qwen-7B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use zhuyaoyu/CodeV-R1-RL-Qwen-7B with Docker Model Runner:
```
docker model run hf.co/zhuyaoyu/CodeV-R1-RL-Qwen-7B
```

zhuyaoyu commited on Jun 10, 2025

Commit

58e62be

verified ·

1 Parent(s): 8a9063b

Update README.md

Browse files

Files changed (1) hide show

README.md +12 -5

README.md CHANGED Viewed

@@ -11,6 +11,17 @@ pipeline_tag: text-generation
 [Project page](https://iprc-dip.github.io/CodeV-R1)
 ### 1. Introduction
 Large language models (LLMs) trained via reinforcement learning with verifiable reward (RLVR) have achieved breakthroughs on tasks with explicit, automatable verification, such as software programming and mathematical problems. Extending RLVR to electronic design automation (EDA), especially automatically generating hardware description languages (HDLs) like Verilog from natural-language (NL) specifications, however, poses three key challenges: the lack of automated and accurate verification environments, the scarcity of high‐quality NL–code pairs, and the prohibitive computation cost of RLVR.
@@ -66,14 +77,10 @@ Our evaluation encompasses Verilog benchmarks, including VerilogEval and RTLLM.
 | **CodeV-R1-distill (ours)** | 7B          | Verilog RTL | 56.2%     |
 | **CodeV-R1 (ours)**         | 7B          | Verilog RTL | **72.9%** |
-We also plot the results for RTLLM v1.1, including pass rate against model size and test-time scaling under different token/FLOPs budgets.
 <div style="display: flex; gap: 10px;">
  <img src="./assets/rtllm_acc_vs_model_size.png" alt="RTLLM TTS Results" width="1200">
 </div>
-<div style="display: flex; gap: 10px;">
- <img src="./assets/rtllm_tts.png" alt="RTLLM TTS Results" width="500">
- <img src="./assets/rtllm_tts_flops.png" alt="RTLLM TTS FLOPs Results" width="500">
-</div>
 ### 4. Usage

 [Project page](https://iprc-dip.github.io/CodeV-R1)
+<div class="figure-container" style="display: flex; flex-direction: column; gap: 15px; max-width: 850px;">
+  <div style="display: flex; gap: 10px; justify-content: center;">
+    <img src="./assets/rtllm_tts.png" alt="RTLLM TTS Results" width="400">
+    <img src="./assets/rtllm_tts_flops.png" alt="RTLLM TTS FLOPs Results" width="400">
+  </div>
+  <figcaption class="caption mt-3 has-text-centered is-size-7 has-text-grey">
+    Test-time scaling curves. Left: Inference time as a function of token length. Right: Inference time vs. estimated FLOPs consumption.
+    When measured by FLOPs consumption, our model achieves better results with fewer computational resources than DeepSeek-R1, highlighting its superior efficiency.
+  </figcaption>
+</div>
 ### 1. Introduction
 Large language models (LLMs) trained via reinforcement learning with verifiable reward (RLVR) have achieved breakthroughs on tasks with explicit, automatable verification, such as software programming and mathematical problems. Extending RLVR to electronic design automation (EDA), especially automatically generating hardware description languages (HDLs) like Verilog from natural-language (NL) specifications, however, poses three key challenges: the lack of automated and accurate verification environments, the scarcity of high‐quality NL–code pairs, and the prohibitive computation cost of RLVR.
 | **CodeV-R1-distill (ours)** | 7B          | Verilog RTL | 56.2%     |
 | **CodeV-R1 (ours)**         | 7B          | Verilog RTL | **72.9%** |
+For RTLLM v1.1, we also plot results showing pass rate against model size.
 <div style="display: flex; gap: 10px;">
  <img src="./assets/rtllm_acc_vs_model_size.png" alt="RTLLM TTS Results" width="1200">
 </div>
 ### 4. Usage