Instructions to use thunlp/LLaMA3-Instruct-8B-FR-Spec with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use thunlp/LLaMA3-Instruct-8B-FR-Spec with Transformers:

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

pipe = pipeline("text-generation", model="thunlp/LLaMA3-Instruct-8B-FR-Spec")

# Load model directly
from transformers import AutoTokenizer, LlamaForCausalLMEagle

tokenizer = AutoTokenizer.from_pretrained("thunlp/LLaMA3-Instruct-8B-FR-Spec")
model = LlamaForCausalLMEagle.from_pretrained("thunlp/LLaMA3-Instruct-8B-FR-Spec")

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use thunlp/LLaMA3-Instruct-8B-FR-Spec with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "thunlp/LLaMA3-Instruct-8B-FR-Spec"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "thunlp/LLaMA3-Instruct-8B-FR-Spec",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/thunlp/LLaMA3-Instruct-8B-FR-Spec

SGLang

How to use thunlp/LLaMA3-Instruct-8B-FR-Spec 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 "thunlp/LLaMA3-Instruct-8B-FR-Spec" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "thunlp/LLaMA3-Instruct-8B-FR-Spec",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

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 "thunlp/LLaMA3-Instruct-8B-FR-Spec" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "thunlp/LLaMA3-Instruct-8B-FR-Spec",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use thunlp/LLaMA3-Instruct-8B-FR-Spec with Docker Model Runner:
```
docker model run hf.co/thunlp/LLaMA3-Instruct-8B-FR-Spec
```

Token frequency statistics based on SlimPajama-627B, used for FR-Spec (https://arxiv.org/abs/2502.14856), see more at https://github.com/thunlp/FR-Spec.

freq_32768.pt can be loaded by torch.load(), and it is a list of high-frequency tokens.

config.json and pytorch_model.bin are the same as https://huggingface.co/yuhuili/EAGLE-LLaMA3-Instruct-8B, and can be downloaded from their repo.

Downloads last month: 1,073

Model tree for thunlp/LLaMA3-Instruct-8B-FR-Spec

Base model

meta-llama/Meta-Llama-3-8B-Instruct

Finetuned

(1095)

this model

Paper for thunlp/LLaMA3-Instruct-8B-FR-Spec

FR-Spec: Accelerating Large-Vocabulary Language Models via Frequency-Ranked Speculative Sampling

Paper • 2502.14856 • Published Feb 20, 2025 • 8