Two Stones Hit One Bird: Bilevel Positional Encoding for Better Length Extrapolation
Paper • 2401.16421 • Published
How to use hzy00/ALiBi-151M with Transformers:
# Use a pipeline as a high-level helper
from transformers import pipeline
pipe = pipeline("text-generation", model="hzy00/ALiBi-151M") # Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM
tokenizer = AutoTokenizer.from_pretrained("hzy00/ALiBi-151M")
model = AutoModelForCausalLM.from_pretrained("hzy00/ALiBi-151M")How to use hzy00/ALiBi-151M with vLLM:
# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "hzy00/ALiBi-151M"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
-H "Content-Type: application/json" \
--data '{
"model": "hzy00/ALiBi-151M",
"prompt": "Once upon a time,",
"max_tokens": 512,
"temperature": 0.5
}'docker model run hf.co/hzy00/ALiBi-151M
How to use hzy00/ALiBi-151M with SGLang:
# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
--model-path "hzy00/ALiBi-151M" \
--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": "hzy00/ALiBi-151M",
"prompt": "Once upon a time,",
"max_tokens": 512,
"temperature": 0.5
}'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 "hzy00/ALiBi-151M" \
--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": "hzy00/ALiBi-151M",
"prompt": "Once upon a time,",
"max_tokens": 512,
"temperature": 0.5
}'How to use hzy00/ALiBi-151M with Docker Model Runner:
docker model run hf.co/hzy00/ALiBi-151M
For more information, please refer to our paper and GitHub repository.
Paper: Two Stones Hit One Bird: Bilevel Positional Encoding for Better Length Extrapolation
GitHub: BiPE
Authors: Zhenyu He*, Guhao Feng*, Shengjie Luo*, Kai Yang, Di He, Jingjing Xu, Zhi Zhang, Hongxia Yang, Liwei Wang