Instructions to use yuyijiong/Llama3-8B-Chinese-Chat-32k with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use yuyijiong/Llama3-8B-Chinese-Chat-32k with Transformers:

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

pipe = pipeline("text-generation", model="yuyijiong/Llama3-8B-Chinese-Chat-32k")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("yuyijiong/Llama3-8B-Chinese-Chat-32k")
model = AutoModelForCausalLM.from_pretrained("yuyijiong/Llama3-8B-Chinese-Chat-32k", 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 yuyijiong/Llama3-8B-Chinese-Chat-32k with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "yuyijiong/Llama3-8B-Chinese-Chat-32k"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "yuyijiong/Llama3-8B-Chinese-Chat-32k",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/yuyijiong/Llama3-8B-Chinese-Chat-32k

SGLang

How to use yuyijiong/Llama3-8B-Chinese-Chat-32k 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 "yuyijiong/Llama3-8B-Chinese-Chat-32k" \
    --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": "yuyijiong/Llama3-8B-Chinese-Chat-32k",
		"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 "yuyijiong/Llama3-8B-Chinese-Chat-32k" \
        --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": "yuyijiong/Llama3-8B-Chinese-Chat-32k",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use yuyijiong/Llama3-8B-Chinese-Chat-32k with Docker Model Runner:
```
docker model run hf.co/yuyijiong/Llama3-8B-Chinese-Chat-32k
```

Llama3-8b-chinese-chat-32k

训练方式

使用 NTK-aware 方法扩展上下文长度至 32k
以 shenzhi-wang/Llama3-8B-Chinese-Chat 为基础在 Long-Instruction-with-Paraphrasing 数据集上，使用 QLora 微调 1 epoch。

使用方法

和原版相同

from transformers import AutoTokenizer, AutoModelForCausalLM

model_id = "yuyijiong/Llama3-8B-Chinese-Chat-32k"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id, torch_dtype="auto", device_map="auto"
)

messages = [
    {"role": "user", "content": "写一首诗吧"},
]

input_ids = tokenizer.apply_chat_template(
    messages, add_generation_prompt=True, return_tensors="pt"
).to(model.device)

outputs = model.generate(
    input_ids,
    max_new_tokens=32768,
    do_sample=True,
    temperature=0.6,
    top_p=0.9,
)
response = outputs[0][input_ids.shape[-1]:]
print(tokenizer.decode(response, skip_special_tokens=True))

Long-Context Performance

相比原始版本，拥有更强的长上下文能力

LongBench (en)

model	hotpotqa	multifieldqa_en	passage_retrieval_en	qmsum	trec
llama3-8b-chinese-chat	45.88	50.56	68.00	22.52	73.00
llama3-8b-chinese-chat-32k	47.64	49.98	100.00	25.13	75.0

Longbench (zh)

model	dureader	multifieldqa_zh	passage_retrieval_zh	vcsum	lsht
llama3-8b-chinese-chat	29.08	58.4	93.5	14.61	28.25
llama3-8b-chinese-chat-32k	32.31	58.66	82.5	16.15	38.5

Downloads last month: 7

Safetensors

Model size

8B params

Tensor type

BF16

Dataset used to train yuyijiong/Llama3-8B-Chinese-Chat-32k

Collection including yuyijiong/Llama3-8B-Chinese-Chat-32k

train_with_paraphrasing

Collection

[long-context models trained with "original text paraphrasing" dataset](https://github.com/yuyijiong/train_with_paraphrasing) • 4 items • Updated Jun 28, 2024

Paper for yuyijiong/Llama3-8B-Chinese-Chat-32k

"Paraphrasing The Original Text" Makes High Accuracy Long-Context QA

Paper • 2312.11193 • Published Dec 18, 2023