Instructions to use sisizhang/klear-agentforge-8b-ppo-step17 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use sisizhang/klear-agentforge-8b-ppo-step17 with Transformers:

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

pipe = pipeline("text-generation", model="sisizhang/klear-agentforge-8b-ppo-step17")
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("sisizhang/klear-agentforge-8b-ppo-step17")
model = AutoModelForCausalLM.from_pretrained("sisizhang/klear-agentforge-8b-ppo-step17", 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 sisizhang/klear-agentforge-8b-ppo-step17 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "sisizhang/klear-agentforge-8b-ppo-step17"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "sisizhang/klear-agentforge-8b-ppo-step17",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/sisizhang/klear-agentforge-8b-ppo-step17

SGLang

How to use sisizhang/klear-agentforge-8b-ppo-step17 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 "sisizhang/klear-agentforge-8b-ppo-step17" \
    --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": "sisizhang/klear-agentforge-8b-ppo-step17",
		"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 "sisizhang/klear-agentforge-8b-ppo-step17" \
        --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": "sisizhang/klear-agentforge-8b-ppo-step17",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use sisizhang/klear-agentforge-8b-ppo-step17 with Docker Model Runner:
```
docker model run hf.co/sisizhang/klear-agentforge-8b-ppo-step17
```

面向代码智能体的 PPO 训练 · RL 模型 / PPO for Code Agents · RL Model

中文

基于 Qwen3-8B 的 Kwai-Klear/Klear-AgentForge-8B-SFT，使用带过程奖励的 PPO 训练得到。本仓库发布最终选定的 step 17 完整合并推理权重、配置、tokenizer 和 chat template。

SWE-bench 500 题效果

复现 SFT 模型：39.8%（199/500）
最终 RL 模型：42.8%（214/500），提升 3.0 个百分点。

模型用于 mini-swe-agent-plus 的长程代码任务。训练样本见公开数据集，价值模型见Critic 模型，项目介绍见Space。

支持该项目

数据、开发、GPU和CPU等训练的费用均由我自行承担。如果您觉得这个模型对您有用，并希望资助更大的数据集和未来的开源版本，可以给我点赞，留言。

English

Trained from the Qwen3-8B-based Kwai-Klear/Klear-AgentForge-8B-SFT with process-reward PPO. This repository provides the selected step 17 merged inference weights, configuration, tokenizer, and chat template.

Results on 500 SWE-bench tasks

Reproduced SFT model: 39.8% (199/500).
Final RL model: 42.8% (214/500), up 3.0 percentage points.

The model targets long-horizon code tasks in mini-swe-agent-plus. See the task dataset, Critic model, and project Space.

Support This Project

I personally cover the costs of data, development, GPU and CPU training. If you find this model useful and would like to support larger datasets and future open-source releases, please give it a like and leave a comment.

Downloads last month: 391

Safetensors

Model size

8B params

Tensor type

BF16

Model tree for sisizhang/klear-agentforge-8b-ppo-step17

Base model

Kwai-Klear/Klear-AgentForge-8B-SFT

Finetuned

(2)

this model

sisizhang
/

klear-agentforge-8b-ppo-step17