Instructions to use Q1ngMang/LFM2.5-230M-PT-zh_CN with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Q1ngMang/LFM2.5-230M-PT-zh_CN with Transformers:

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

pipe = pipeline("text-generation", model="Q1ngMang/LFM2.5-230M-PT-zh_CN")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("Q1ngMang/LFM2.5-230M-PT-zh_CN")
model = AutoModelForCausalLM.from_pretrained("Q1ngMang/LFM2.5-230M-PT-zh_CN", 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]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use Q1ngMang/LFM2.5-230M-PT-zh_CN with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Q1ngMang/LFM2.5-230M-PT-zh_CN"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Q1ngMang/LFM2.5-230M-PT-zh_CN",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Q1ngMang/LFM2.5-230M-PT-zh_CN

SGLang

How to use Q1ngMang/LFM2.5-230M-PT-zh_CN 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 "Q1ngMang/LFM2.5-230M-PT-zh_CN" \
    --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": "Q1ngMang/LFM2.5-230M-PT-zh_CN",
		"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 "Q1ngMang/LFM2.5-230M-PT-zh_CN" \
        --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": "Q1ngMang/LFM2.5-230M-PT-zh_CN",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use Q1ngMang/LFM2.5-230M-PT-zh_CN with Docker Model Runner:
```
docker model run hf.co/Q1ngMang/LFM2.5-230M-PT-zh_CN
```

LFM2.5-350M-PT-zh_CN

基于 LiquidAI/LFM2.5-230M 进行 Pre-Training 的模型。
数据集 0xDing/wikipedia-cn-20230720-filtered

协议冲突

LFM 1.0 规定年收入小于1000万美元的公司的免费商用，收入大于1000万美元需要向LiquidAI购买商业许可证，而 CC BY-SA 3.0 规定可以商用，因此存在冲突。
该模型优先使用 CC BY-SA 3.0，但并不代表收入大于1000万美元不需要向LiquidAI购买商业许可证。

便秘数据

平台：LLaMA Factory
GPU：NVIDIA Tesla P100 SXM2

便秘参数

项	值
微调方法	LoRA
量化等级	4
量化方法	bnb
对话模板	lfm2
训练阶段	Pre-Training
学习率	4.5e-4
训练轮数	1.0
计算类型	pure_bf16
截断长度	1968
批处理大小	7
梯度累积	147
验证集比例	0.05
序列打包	True
额外参数	{"optim": "paged_adamw_8bit"}
LoRA秩	16
LoRA缩放系数	32

便秘结果

项	值
输入Token	158,319,896
计算量	144,139,704GF
训练损失	3.7876
训练耗时	12:35:07.48
每秒训练样本数	1.776
每秒训练步数	0.002
验证损失	3.6571
验证困惑度	38.7494
验证耗时	0:16:13.08
每秒验证样本数	4.409
每秒验证步数	0.63

Downloads last month: 21

Safetensors

Model size

0.2B params

Tensor type

BF16

Model tree for Q1ngMang/LFM2.5-230M-PT-zh_CN

Base model

LiquidAI/LFM2.5-230M-Base

Finetuned

LiquidAI/LFM2.5-230M

Finetuned

(29)

this model

Q1ngMang
/

LFM2.5-230M-PT-zh_CN