Instructions to use zhangux/TinyChat-0.1B-sft with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use zhangux/TinyChat-0.1B-sft with Transformers:

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

pipe = pipeline("text-generation", model="zhangux/TinyChat-0.1B-sft", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("zhangux/TinyChat-0.1B-sft", trust_remote_code=True, device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use zhangux/TinyChat-0.1B-sft with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "zhangux/TinyChat-0.1B-sft"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "zhangux/TinyChat-0.1B-sft",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/zhangux/TinyChat-0.1B-sft

SGLang

How to use zhangux/TinyChat-0.1B-sft 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 "zhangux/TinyChat-0.1B-sft" \
    --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": "zhangux/TinyChat-0.1B-sft",
		"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 "zhangux/TinyChat-0.1B-sft" \
        --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": "zhangux/TinyChat-0.1B-sft",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use zhangux/TinyChat-0.1B-sft with Docker Model Runner:
```
docker model run hf.co/zhangux/TinyChat-0.1B-sft
```

TinyChat-0.1B-sft

这是sft版本，后续的RLHF版本会基于此模型进行微调。

模型特性

参数量: 约 0.1B (112M)
架构: Decoder-only Transformer
注意力机制: Grouped Query Attention (GQA)
位置编码: Rotary Position Embedding (RoPE)
归一化: RMSNorm
激活函数: SiLU (Swish)
词汇表大小: 6,400
最大序列长度: 512 tokens

模型配置

- Layers: 12
- Hidden Size: 768
- Attention Heads: 16 (Query) / 4 (Key-Value)
- FFN Hidden Size: 3,072
- Dropout: 0.1

使用方法

安装依赖

pip install torch==2.4.0 transformers==4.46.3

使用示例

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

# 加载模型和分词器
model = AutoModelForCausalLM.from_pretrained(
    "zhangux/TinyChat-0.1B-sft",
    trust_remote_code=True
)
tokenizer = AutoTokenizer.from_pretrained(
    "zhangux/TinyChat-0.1B-sft"
)

prompt = "今天天气怎么样？"
messages = [{"role": "user", "content": prompt}]
text = tokenizer.apply_chat_template(
            messages,
            tokenize=False,
            add_generation_prompt=True
        )

inputs = tokenizer(text, return_tensors="pt")
input_ids = inputs.input_ids
output_ids = model.generate(
        input_ids=input_ids,
        do_sample=True,
        pad_token_id=tokenizer.pad_token_id,
        eos_token_id=tokenizer.eos_token_id,
        use_cache=True # 如果这里出现了 kv_cache 相关的错误，尝试 use_cache=False
    )

generated_ids = output_ids[0][len(input_ids[0]):]
response = tokenizer.decode(generated_ids, skip_special_tokens=True)
print(response)

参考结果

今天的天气状况：晴朗，无云，无雨，风力较小。建议出门时带上太阳镜和防晒霜。

loss & accuracy

Downloads last month: 167

Safetensors

Model size

0.1B params

Tensor type

F32