yinqiwen / lmsf

Rust LLM Serving Framework

Features

• Paged Attention
• Continuous Batch
• Quantization
  • awq
  • squeezellm
• Models
  • llama
  • gemma
  • chatglm

Getting Started

Examples

$ cargo run --release --example llm_engine_example -- --model <llma model dir> --gpu-memory-utilization 0.95 --block-size 8 --max-model-len 1024

API Server

$ cargo build --release
$ ./target/release/entrypoints --model <llma model dir> --gpu-memory-utilization 0.95 --block-size 8 --max-model-len 1024 --host 0.0.0.0 --port 8000