Instructions to use GnLOLot/MiniCPM5-2B-Claude-Fable5-1-Thinking-Agentic with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use GnLOLot/MiniCPM5-2B-Claude-Fable5-1-Thinking-Agentic with Transformers:

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

pipe = pipeline("text-generation", model="GnLOLot/MiniCPM5-2B-Claude-Fable5-1-Thinking-Agentic")
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("GnLOLot/MiniCPM5-2B-Claude-Fable5-1-Thinking-Agentic")
model = AutoModelForCausalLM.from_pretrained("GnLOLot/MiniCPM5-2B-Claude-Fable5-1-Thinking-Agentic", 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 GnLOLot/MiniCPM5-2B-Claude-Fable5-1-Thinking-Agentic with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "GnLOLot/MiniCPM5-2B-Claude-Fable5-1-Thinking-Agentic"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "GnLOLot/MiniCPM5-2B-Claude-Fable5-1-Thinking-Agentic",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/GnLOLot/MiniCPM5-2B-Claude-Fable5-1-Thinking-Agentic

SGLang

How to use GnLOLot/MiniCPM5-2B-Claude-Fable5-1-Thinking-Agentic 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 "GnLOLot/MiniCPM5-2B-Claude-Fable5-1-Thinking-Agentic" \
    --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": "GnLOLot/MiniCPM5-2B-Claude-Fable5-1-Thinking-Agentic",
		"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 "GnLOLot/MiniCPM5-2B-Claude-Fable5-1-Thinking-Agentic" \
        --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": "GnLOLot/MiniCPM5-2B-Claude-Fable5-1-Thinking-Agentic",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use GnLOLot/MiniCPM5-2B-Claude-Fable5-1-Thinking-Agentic with Docker Model Runner:
```
docker model run hf.co/GnLOLot/MiniCPM5-2B-Claude-Fable5-1-Thinking-Agentic
```

MiniCPM5-2B-Claude-Fable5-1-Thinking-Agentic / README_zh.md

GnLOLot

Initial release: MiniCPM5-2B-Claude-Fable5-1-Thinking-Agentic

934bcdf verified 15 days ago

preview code

Raw History Blame

5.04 kB

MiniCPM5-2B-Claude-Fable5-1-Thinking-Agentic

GGUF 量化版本（用于本地部署）：**MiniCPM5-2B-Claude-Fable5-1-Thinking-Agentic-GGUF**

English

MiniCPM5-2B-Claude-Fable5-1-Thinking-Agentic 是一个基于 openbmb/MiniCPM5-2B 的紧凑型 2B 思维链（Thinking） 语言模型。使用 Claude 数据进行微调，重点增强了 Agent 工具调用 / 函数调用、代码生成 和 指令遵循 能力，同时保留了 MiniCPM5 原生的 Thinking 聊天模板和 XML 工具调用格式。

概览

项目	详情
基座模型	openbmb/MiniCPM5-2B（2B 稠密 Llama 架构）
后训练数据	Claude 数据
核心能力	Agent 工具调用、代码生成、指令遵循、思维链推理
聊天格式	MiniCPM5 原生 Thinking 模板，支持可选的思维链输出
上下文长度	128K（`max_position_embeddings = 131072`）
精度	bfloat16
部署	单卡友好，适合边缘设备 / 本地部署

评测结果

ClawBench（Agent 代码能力）

模型	QwenClawBench	WildClawBench
MiniCPM5-2B（基座，仅 RL）	42.11	23.19
MiniCPM5-2B-Claude-Fable5-1-Thinking-Agentic	44.56（+2.45）	24.32（+1.13）

ClawBench 评测 Agent 代码能力——模型自主使用工具、浏览代码库、完成多步软件工程任务的综合能力。QwenClawBench 使用结构化代码场景；WildClawBench 测试多样化的真实任务。

更多评测成绩（BFCL、SWE-bench、Tau-Bench 等）补充中。

能力

Agent 工具调用 — 基于 MiniCPM5 原生格式的可靠 XML / 函数调用风格工具使用，适配多步 Agent 工作流
代码生成 — 代码编写、调试及软件工程任务
指令遵循 — 可靠地遵循用户指令和结构化约束
思维链模式 — 通过 MiniCPM5 聊天模板实现链式推理
长上下文 — 最高支持 128K tokens（131,072 tokens）

快速开始

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_id = "GnLOLot/MiniCPM5-2B-Claude-Fable5-1-Thinking-Agentic"

tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    trust_remote_code=True,
    torch_dtype=torch.bfloat16,
    device_map="auto",
)

messages = [{"role": "user", "content": "写一个合并两个有序列表的 Python 函数。"}]
text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(text, return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=512, do_sample=False)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True))

工具调用示例

tools = [
    {
        "type": "function",
        "function": {
            "name": "get_weather",
            "description": "获取指定城市的当前天气。",
            "parameters": {
                "type": "object",
                "properties": {
                    "city": {"type": "string", "description": "城市名称"}
                },
                "required": ["city"]
            }
        }
    }
]

messages = [
    {"role": "user", "content": "北京今天天气怎么样？"}
]

text = tokenizer.apply_chat_template(
    messages, tools=tools, tokenize=False, add_generation_prompt=True
)
inputs = tokenizer(text, return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=256, do_sample=False)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True))

采样建议

继承自 openbmb/MiniCPM5-2B：

场景	参数
默认	`temperature=1.0, top_p=0.95, min_p=0.0`
如遇重复输出	`temperature=1.0, top_p=0.95, min_p=0.0, repetition_penalty=1.05`

本模型为 纯思考模式（Thinking-only）——思维链推理始终开启。

不同推理框架对采样参数的支持有所差异，请参考所用框架的文档。

局限性

思维链输出 — 模型可能在最终回答前输出推理过程，下游应用可在展示前过滤
2B 规模 — 针对轻量级本地部署优化，非前沿级通用推理

来源与许可

基于 Apache-2.0 发布，继承自 MiniCPM5-2B。

致谢

基座模型：OpenBMB / MiniCPM5-2B
GGUF 转换：llama.cpp