Update README.md

README.md

@@ -6,6 +6,8 @@
 
 After installation, you can access to `hip` package from any project. `hip` is code name of HiP attention.
 
+[vLLM Implementation](https://github.com/DeepAuto-AI/vllm)
+
 ## TL;DR
 
 **We provide OpenAI Compatible API server built with vLLM and HiP attention!**
@@ -284,4 +286,4 @@ python examples/openai_chat_image_client.py --image-file="images/cherry_blossom.
 ## Example training command
 ```bash
 OMP_NUM_THREADS=16 CUDA_VISIBLE_DEVICES=0,1,2,3 PYTHONPATH=. accelerate launch --num_processes=4 --main_process_port 29501 hip/trainer/hip_trainer_hf.py --method hip --block_size_q 32 --block_size_k 2 --k 512 --lora_r 256 --dataset openwebtext --dense_layers 4 --name bs16_warmup10_dq2k --dense_queries 2048 --seq_len 32768 --disable_kd --sparsity_reg 0.01 --gradient_accumulation_steps 4 --warmup_steps 10 --model giraffe13b --using_deepspeed
-```
+```