Remove enforce-eager to enable HPU graphs for better vLLM perf (opea-project#954)

* remove enforce-eager to enable HPU graphs

Signed-off-by: Wang, Kai Lawrence <kai.lawrence.wang@intel.com>

* Increase the llm max timeout in ci for fully warmup

Signed-off-by: Wang, Kai Lawrence <kai.lawrence.wang@intel.com>

---------

Signed-off-by: Wang, Kai Lawrence <kai.lawrence.wang@intel.com>

Author: wangkl2

comps/llms/faq-generation/vllm/langchain/docker_compose_llm.yaml

@@ -23,7 +23,7 @@ services:
     cap_add:
       - SYS_NICE
     ipc: host
-    command: --enforce-eager --model $LLM_MODEL_ID --tensor-parallel-size 1 --host 0.0.0.0 --port 80
+    command: --model $LLM_MODEL_ID --tensor-parallel-size 1 --host 0.0.0.0 --port 80
   llm:
     image: opea/llm-faqgen-vllm:latest
     container_name: llm-faqgen-server

comps/llms/summarization/vllm/langchain/docker_compose_llm.yaml

@@ -23,7 +23,7 @@ services:
     cap_add:
       - SYS_NICE
     ipc: host
-    command: --enforce-eager --model $LLM_MODEL_ID --tensor-parallel-size 1 --host 0.0.0.0 --port 80
+    command: --model $LLM_MODEL_ID --tensor-parallel-size 1 --host 0.0.0.0 --port 80
   llm:
     image: opea/llm-docsum-vllm:latest
     container_name: llm-docsum-vllm-server

comps/llms/text-generation/vllm/langchain/dependency/launch_vllm_service.sh

@@ -38,7 +38,7 @@ volume=$PWD/data
 
 # Build the Docker run command based on hardware mode
 if [ "$hw_mode" = "hpu" ]; then
-    docker run -d --rm --runtime=habana --name="vllm-service" -p $port_number:80 -e HABANA_VISIBLE_DEVICES=all -e OMPI_MCA_btl_vader_single_copy_mechanism=none --cap-add=sys_nice --ipc=host -e HTTPS_PROXY=$https_proxy -e HTTP_PROXY=$https_proxy -e HF_TOKEN=${HF_TOKEN} opea/vllm-gaudi:latest --enforce-eager --model $model_name  --tensor-parallel-size $parallel_number --host 0.0.0.0 --port 80 --block-size $block_size --max-num-seqs  $max_num_seqs --max-seq_len-to-capture $max_seq_len_to_capture
+    docker run -d --rm --runtime=habana --name="vllm-service" -p $port_number:80 -e HABANA_VISIBLE_DEVICES=all -e OMPI_MCA_btl_vader_single_copy_mechanism=none --cap-add=sys_nice --ipc=host -e HTTPS_PROXY=$https_proxy -e HTTP_PROXY=$https_proxy -e HF_TOKEN=${HF_TOKEN} opea/vllm-gaudi:latest --model $model_name  --tensor-parallel-size $parallel_number --host 0.0.0.0 --port 80 --block-size $block_size --max-num-seqs  $max_num_seqs --max-seq_len-to-capture $max_seq_len_to_capture
 else
     docker run -d --rm --name="vllm-service" -p $port_number:80 --network=host -v $volume:/data -e HTTPS_PROXY=$https_proxy -e HTTP_PROXY=$https_proxy -e HF_TOKEN=${HF_TOKEN} -e VLLM_CPU_KVCACHE_SPACE=40 opea/vllm-cpu:latest --model $model_name --host 0.0.0.0 --port 80
 fi

comps/llms/text-generation/vllm/langchain/docker_compose_llm.yaml

@@ -23,7 +23,7 @@ services:
     cap_add:
       - SYS_NICE
     ipc: host
-    command: --enforce-eager --model $LLM_MODEL --tensor-parallel-size 1 --host 0.0.0.0 --port 80
+    command: --model $LLM_MODEL --tensor-parallel-size 1 --host 0.0.0.0 --port 80
   llm:
     image: opea/llm-vllm:latest
     container_name: llm-vllm-gaudi-server

comps/llms/text-generation/vllm/llama_index/dependency/launch_vllm_service.sh

@@ -38,7 +38,7 @@ volume=$PWD/data
 
 # Build the Docker run command based on hardware mode
 if [ "$hw_mode" = "hpu" ]; then
-    docker run -d --rm --runtime=habana --name="vllm-service" -p $port_number:80 -e HABANA_VISIBLE_DEVICES=all -e OMPI_MCA_btl_vader_single_copy_mechanism=none --cap-add=sys_nice --ipc=host -e HTTPS_PROXY=$https_proxy -e HTTP_PROXY=$https_proxy -e HF_TOKEN=${HUGGINGFACEHUB_API_TOKEN} opea/vllm-gaudi:latest --enforce-eager --model $model_name  --tensor-parallel-size $parallel_number --host 0.0.0.0 --port 80 --block-size $block_size --max-num-seqs  $max_num_seqs --max-seq_len-to-capture $max_seq_len_to_capture
+    docker run -d --rm --runtime=habana --name="vllm-service" -p $port_number:80 -e HABANA_VISIBLE_DEVICES=all -e OMPI_MCA_btl_vader_single_copy_mechanism=none --cap-add=sys_nice --ipc=host -e HTTPS_PROXY=$https_proxy -e HTTP_PROXY=$https_proxy -e HF_TOKEN=${HUGGINGFACEHUB_API_TOKEN} opea/vllm-gaudi:latest --model $model_name  --tensor-parallel-size $parallel_number --host 0.0.0.0 --port 80 --block-size $block_size --max-num-seqs  $max_num_seqs --max-seq_len-to-capture $max_seq_len_to_capture
 else
     docker run -d --rm --name="vllm-service" -p $port_number:80 --network=host -v $volume:/data -e HTTPS_PROXY=$https_proxy -e HTTP_PROXY=$https_proxy -e HF_TOKEN=${HUGGINGFACEHUB_API_TOKEN} -e VLLM_CPU_KVCACHE_SPACE=40 opea/vllm-cpu:latest --model $model_name --host 0.0.0.0 --port 80
 fi

comps/llms/text-generation/vllm/llama_index/docker_compose_llm.yaml

@@ -23,7 +23,7 @@ services:
     cap_add:
       - SYS_NICE
     ipc: host
-    command: --enforce-eager --model $LLM_MODEL --tensor-parallel-size 1 --host 0.0.0.0 --port 80
+    command: --model $LLM_MODEL --tensor-parallel-size 1 --host 0.0.0.0 --port 80
   llm:
     image: opea/llm-vllm-llamaindex:latest
     container_name: llm-vllm-gaudi-server

tests/llms/test_llms_text-generation_vllm_langchain_on_intel_hpu.sh

@@ -48,7 +48,7 @@ function start_service() {
         --ipc=host \
         -e HF_TOKEN=${HUGGINGFACEHUB_API_TOKEN} \
         opea/vllm-gaudi:comps \
-        --enforce-eager --model $LLM_MODEL  --tensor-parallel-size 1 --host 0.0.0.0 --port 80 --block-size 128 --max-num-seqs 256 --max-seq_len-to-capture 2048
+        --model $LLM_MODEL  --tensor-parallel-size 1 --host 0.0.0.0 --port 80 --block-size 128 --max-num-seqs 256 --max-seq_len-to-capture 2048
 
     export vLLM_ENDPOINT="http://${ip_address}:${port_number}"
     docker run -d --rm \
@@ -62,7 +62,7 @@ function start_service() {
 
     # check whether vllm ray is fully ready
     n=0
-    until [[ "$n" -ge 120 ]] || [[ $ready == true ]]; do
+    until [[ "$n" -ge 160 ]] || [[ $ready == true ]]; do
         docker logs test-comps-vllm-service > ${WORKPATH}/tests/test-comps-vllm-service.log
         n=$((n+1))
         if grep -q throughput ${WORKPATH}/tests/test-comps-vllm-service.log; then

tests/llms/test_llms_text-generation_vllm_llamaindex_on_intel_hpu.sh

@@ -48,7 +48,7 @@ function start_service() {
         --ipc=host \
         -e HF_TOKEN=${HUGGINGFACEHUB_API_TOKEN} \
         opea/vllm-gaudi:comps \
-        --enforce-eager --model $LLM_MODEL  --tensor-parallel-size 1 --host 0.0.0.0 --port 80 --block-size 128 --max-num-seqs 256 --max-seq_len-to-capture 2048
+        --model $LLM_MODEL  --tensor-parallel-size 1 --host 0.0.0.0 --port 80 --block-size 128 --max-num-seqs 256 --max-seq_len-to-capture 2048
 
     export vLLM_ENDPOINT="http://${ip_address}:${port_number}"
     docker run -d --rm \
@@ -62,7 +62,7 @@ function start_service() {
 
     # check whether vllm ray is fully ready
     n=0
-    until [[ "$n" -ge 120 ]] || [[ $ready == true ]]; do
+    until [[ "$n" -ge 160 ]] || [[ $ready == true ]]; do
         docker logs test-comps-vllm-service > ${WORKPATH}/tests/test-comps-vllm-service.log
         n=$((n+1))
         if grep -q throughput ${WORKPATH}/tests/test-comps-vllm-service.log; then