Skip to content

test: L0_orca_trtllm fixed #8191

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 3 commits into
base: main
Choose a base branch
from
Open

test: L0_orca_trtllm fixed #8191

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
bf88f4a
Fix L0_orca_trtllm which was broken due to new changes in the trtllm …
indrajit96 Apr 29, 2025
d638f04
Merge branch 'main' into orca_fix
indrajit96 May 15, 2025
5d32dff
Move common functions to common/util.sh
indrajit96 May 15, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
36 changes: 1 addition & 35 deletions qa/L0_orca/test.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -36,7 +36,7 @@ MODEL_NAME="gpt2_tensorrt_llm"
NAME="tensorrt_llm_benchmarking_test"
MODEL_REPOSITORY="$(pwd)/triton_model_repo"
TENSORRTLLM_BACKEND_DIR="/workspace/tensorrtllm_backend"
GPT_DIR="$TENSORRTLLM_BACKEND_DIR/tensorrt_llm/examples/gpt"
GPT_DIR="$TENSORRTLLM_BACKEND_DIR/tensorrt_llm/examples/models/core/gpt"
TOKENIZER_DIR="$GPT_DIR/gpt2"
ENGINES_DIR="${BASE_DIR}/engines/inflight_batcher_llm/${NUM_GPUS}-gpu"
TRITON_DIR=${TRITON_DIR:="/opt/tritonserver"}
Expand All @@ -48,40 +48,6 @@ CLIENT_PY=${BASE_DIR}/orca_http_test.py
CLIENT_LOG="${NAME}_orca_http_test.log"
source ../common/util.sh

function prepare_model_repository {
rm -rf ${MODEL_REPOSITORY} && mkdir ${MODEL_REPOSITORY}
cp -r ${TENSORRTLLM_BACKEND_DIR}/all_models/inflight_batcher_llm/* ${MODEL_REPOSITORY}
rm -rf ${MODEL_REPOSITORY}/tensorrt_llm_bls
mv "${MODEL_REPOSITORY}/ensemble" "${MODEL_REPOSITORY}/${MODEL_NAME}"

replace_config_tags "model_version: -1" "model_version: 1" "${MODEL_REPOSITORY}/${MODEL_NAME}/config.pbtxt"
replace_config_tags '${triton_max_batch_size}' "128" "${MODEL_REPOSITORY}/${MODEL_NAME}/config.pbtxt"
replace_config_tags 'name: "ensemble"' "name: \"$MODEL_NAME\"" "${MODEL_REPOSITORY}/${MODEL_NAME}/config.pbtxt"
replace_config_tags '${logits_datatype}' "TYPE_FP32" "${MODEL_REPOSITORY}/${MODEL_NAME}/config.pbtxt"

replace_config_tags '${triton_max_batch_size}' "128" "${MODEL_REPOSITORY}/preprocessing/config.pbtxt"
replace_config_tags '${preprocessing_instance_count}' '1' "${MODEL_REPOSITORY}/preprocessing/config.pbtxt"
replace_config_tags '${tokenizer_dir}' "${TOKENIZER_DIR}/" "${MODEL_REPOSITORY}/preprocessing/config.pbtxt"
replace_config_tags '${logits_datatype}' "TYPE_FP32" "${MODEL_REPOSITORY}/preprocessing/config.pbtxt"
replace_config_tags '${max_queue_delay_microseconds}' "1000000" "${MODEL_REPOSITORY}/preprocessing/config.pbtxt"
replace_config_tags '${max_queue_size}' "0" "${MODEL_REPOSITORY}/preprocessing/config.pbtxt"

replace_config_tags '${triton_max_batch_size}' "128" "${MODEL_REPOSITORY}/postprocessing/config.pbtxt"
replace_config_tags '${postprocessing_instance_count}' '1' "${MODEL_REPOSITORY}/postprocessing/config.pbtxt"
replace_config_tags '${tokenizer_dir}' "${TOKENIZER_DIR}/" "${MODEL_REPOSITORY}/postprocessing/config.pbtxt"
replace_config_tags '${logits_datatype}' "TYPE_FP32" "${MODEL_REPOSITORY}/postprocessing/config.pbtxt"

replace_config_tags '${triton_max_batch_size}' "128" "${MODEL_REPOSITORY}/tensorrt_llm/config.pbtxt"
replace_config_tags '${decoupled_mode}' 'true' "${MODEL_REPOSITORY}/tensorrt_llm/config.pbtxt"
replace_config_tags '${max_queue_delay_microseconds}' "1000000" "${MODEL_REPOSITORY}/tensorrt_llm/config.pbtxt"
replace_config_tags '${batching_strategy}' 'inflight_fused_batching' "${MODEL_REPOSITORY}/tensorrt_llm/config.pbtxt"
replace_config_tags '${engine_dir}' "${ENGINES_DIR}" "${MODEL_REPOSITORY}/tensorrt_llm/config.pbtxt"
replace_config_tags '${triton_backend}' "tensorrtllm" "${MODEL_REPOSITORY}/tensorrt_llm/config.pbtxt"
replace_config_tags '${max_queue_size}' "0" "${MODEL_REPOSITORY}/tensorrt_llm/config.pbtxt"
replace_config_tags '${logits_datatype}' "TYPE_FP32" "${MODEL_REPOSITORY}/tensorrt_llm/config.pbtxt"
replace_config_tags '${encoder_input_features_data_type}' "TYPE_FP32" "${MODEL_REPOSITORY}/tensorrt_llm/config.pbtxt"
}

# Wait until server health endpoint shows ready. Sets WAIT_RET to 0 on
# success, 1 on failure
function wait_for_server_ready() {
Expand Down
75 changes: 3 additions & 72 deletions qa/L0_perf_tensorrt_llm/test.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,5 +1,5 @@
#!/bin/bash
# Copyright 2024, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
# Copyright 2024-2025, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
#
# Redistribution and use in source and binary forms, with or without
# modification, are permitted provided that the following conditions
Expand Down Expand Up @@ -35,21 +35,15 @@ MODEL_NAME="gpt2_tensorrt_llm"
NAME="tensorrt_llm_benchmarking_test"
MODEL_REPOSITORY="$(pwd)/triton_model_repo"
TENSORRTLLM_BACKEND_DIR="/workspace/tensorrtllm_backend"
GPT_DIR="$TENSORRTLLM_BACKEND_DIR/tensorrt_llm/examples/gpt"
GPT_DIR="$TENSORRTLLM_BACKEND_DIR/tensorrt_llm/examples/models/core/gpt"
TOKENIZER_DIR="$GPT_DIR/gpt2"
ENGINES_DIR="${BASE_DIR}/engines/inflight_batcher_llm/${NUM_GPUS}-gpu"
TRITON_DIR=${TRITON_DIR:="/opt/tritonserver"}
SERVER=${TRITON_DIR}/bin/tritonserver
BACKEND_DIR=${TRITON_DIR}/backends
SERVER_LOG="${NAME}_server.log"
SERVER_TIMEOUT=${SERVER_TIMEOUT:=120}

function clone_tensorrt_llm_backend_repo {
rm -rf $TENSORRTLLM_BACKEND_DIR && mkdir $TENSORRTLLM_BACKEND_DIR
apt-get update && apt-get install git-lfs -y --no-install-recommends
git clone --single-branch --depth=1 -b ${TENSORRTLLM_BACKEND_REPO_TAG} ${TRITON_REPO_ORG}/tensorrtllm_backend.git $TENSORRTLLM_BACKEND_DIR
cd $TENSORRTLLM_BACKEND_DIR && git lfs install && git submodule update --init --recursive
}
source ../common/util.sh

# Update Open MPI to a version compatible with SLURM.
function upgrade_openmpi {
Expand Down Expand Up @@ -95,69 +89,6 @@ function upgrade_openmpi {
mpirun --version
}

function build_gpt2_base_model {
# Download weights from HuggingFace Transformers
cd ${GPT_DIR} && rm -rf gpt2 && git clone https://huggingface.co/gpt2-medium gpt2 && cd gpt2
rm pytorch_model.bin model.safetensors
if ! wget -q https://huggingface.co/gpt2-medium/resolve/main/pytorch_model.bin; then
echo "Downloading pytorch_model.bin failed."
exit 1
fi
cd ${GPT_DIR}

# Convert weights from HF Tranformers to FT format
python3 convert_checkpoint.py --model_dir gpt2 --dtype float16 --tp_size ${NUM_GPUS} --output_dir "./c-model/gpt2/${NUM_GPUS}-gpu/"
cd ${BASE_DIR}
}

function build_gpt2_tensorrt_engine {
# Build TensorRT engines
cd ${GPT_DIR}
trtllm-build --checkpoint_dir "./c-model/gpt2/${NUM_GPUS}-gpu/" \
--gpt_attention_plugin float16 \
--remove_input_padding enable \
--paged_kv_cache enable \
--gemm_plugin float16 \
--workers "${NUM_GPUS}" \
--output_dir "${ENGINES_DIR}"

cd ${BASE_DIR}
}

function replace_config_tags {
tag_to_replace="${1}"
new_value="${2}"
config_file_path="${3}"
sed -i "s|${tag_to_replace}|${new_value}|g" ${config_file_path}
}

function prepare_model_repository {
rm -rf ${MODEL_REPOSITORY} && mkdir ${MODEL_REPOSITORY}
cp -r ${TENSORRTLLM_BACKEND_DIR}/all_models/inflight_batcher_llm/* ${MODEL_REPOSITORY}
rm -rf ${MODEL_REPOSITORY}/tensorrt_llm_bls
mv "${MODEL_REPOSITORY}/ensemble" "${MODEL_REPOSITORY}/${MODEL_NAME}"

replace_config_tags "model_version: -1" "model_version: 1" "${MODEL_REPOSITORY}/${MODEL_NAME}/config.pbtxt"
replace_config_tags '${triton_max_batch_size}' "128" "${MODEL_REPOSITORY}/${MODEL_NAME}/config.pbtxt"
replace_config_tags 'name: "ensemble"' "name: \"$MODEL_NAME\"" "${MODEL_REPOSITORY}/${MODEL_NAME}/config.pbtxt"

replace_config_tags '${triton_max_batch_size}' "128" "${MODEL_REPOSITORY}/preprocessing/config.pbtxt"
replace_config_tags '${preprocessing_instance_count}' '1' "${MODEL_REPOSITORY}/preprocessing/config.pbtxt"
replace_config_tags '${tokenizer_dir}' "${TOKENIZER_DIR}/" "${MODEL_REPOSITORY}/preprocessing/config.pbtxt"

replace_config_tags '${triton_max_batch_size}' "128" "${MODEL_REPOSITORY}/postprocessing/config.pbtxt"
replace_config_tags '${postprocessing_instance_count}' '1' "${MODEL_REPOSITORY}/postprocessing/config.pbtxt"
replace_config_tags '${tokenizer_dir}' "${TOKENIZER_DIR}/" "${MODEL_REPOSITORY}/postprocessing/config.pbtxt"

replace_config_tags '${triton_max_batch_size}' "128" "${MODEL_REPOSITORY}/tensorrt_llm/config.pbtxt"
replace_config_tags '${decoupled_mode}' 'true' "${MODEL_REPOSITORY}/tensorrt_llm/config.pbtxt"
replace_config_tags '${max_queue_delay_microseconds}' "1000000" "${MODEL_REPOSITORY}/tensorrt_llm/config.pbtxt"
replace_config_tags '${batching_strategy}' 'inflight_fused_batching' "${MODEL_REPOSITORY}/tensorrt_llm/config.pbtxt"
replace_config_tags '${engine_dir}' "${ENGINES_DIR}" "${MODEL_REPOSITORY}/tensorrt_llm/config.pbtxt"
replace_config_tags '${triton_backend}' "tensorrtllm" "${MODEL_REPOSITORY}/tensorrt_llm/config.pbtxt"
replace_config_tags '${max_queue_size}' "0" "${MODEL_REPOSITORY}/tensorrt_llm/config.pbtxt"
}

# Wait until server health endpoint shows ready. Sets WAIT_RET to 0 on
# success, 1 on failure
function wait_for_server_ready() {
Expand Down
79 changes: 78 additions & 1 deletion qa/common/util.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,5 +1,5 @@
#!/bin/bash
# Copyright 2018-2024, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
# Copyright 2018-2025, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
#
# Redistribution and use in source and binary forms, with or without
# modification, are permitted provided that the following conditions
Expand Down Expand Up @@ -534,3 +534,80 @@ function deactivate_virtualenv() {
rm -fr venv
fi
}

function replace_config_tags {
tag_to_replace="${1}"
new_value="${2}"
config_file_path="${3}"
sed -i "s|${tag_to_replace}|${new_value}|g" ${config_file_path}
}

function prepare_model_repository {
rm -rf ${MODEL_REPOSITORY} && mkdir ${MODEL_REPOSITORY}
cp -r ${TENSORRTLLM_BACKEND_DIR}/all_models/inflight_batcher_llm/* ${MODEL_REPOSITORY}
rm -rf ${MODEL_REPOSITORY}/tensorrt_llm_bls
mv "${MODEL_REPOSITORY}/ensemble" "${MODEL_REPOSITORY}/${MODEL_NAME}"

replace_config_tags "model_version: -1" "model_version: 1" "${MODEL_REPOSITORY}/${MODEL_NAME}/config.pbtxt"
replace_config_tags '${triton_max_batch_size}' "128" "${MODEL_REPOSITORY}/${MODEL_NAME}/config.pbtxt"
replace_config_tags 'name: "ensemble"' "name: \"$MODEL_NAME\"" "${MODEL_REPOSITORY}/${MODEL_NAME}/config.pbtxt"
replace_config_tags '${logits_datatype}' "TYPE_FP32" "${MODEL_REPOSITORY}/${MODEL_NAME}/config.pbtxt"

replace_config_tags '${triton_max_batch_size}' "128" "${MODEL_REPOSITORY}/preprocessing/config.pbtxt"
replace_config_tags '${preprocessing_instance_count}' '1' "${MODEL_REPOSITORY}/preprocessing/config.pbtxt"
replace_config_tags '${tokenizer_dir}' "${TOKENIZER_DIR}/" "${MODEL_REPOSITORY}/preprocessing/config.pbtxt"
replace_config_tags '${logits_datatype}' "TYPE_FP32" "${MODEL_REPOSITORY}/preprocessing/config.pbtxt"
replace_config_tags '${max_queue_delay_microseconds}' "1000000" "${MODEL_REPOSITORY}/preprocessing/config.pbtxt"
replace_config_tags '${max_queue_size}' "0" "${MODEL_REPOSITORY}/preprocessing/config.pbtxt"

replace_config_tags '${triton_max_batch_size}' "128" "${MODEL_REPOSITORY}/postprocessing/config.pbtxt"
replace_config_tags '${postprocessing_instance_count}' '1' "${MODEL_REPOSITORY}/postprocessing/config.pbtxt"
replace_config_tags '${tokenizer_dir}' "${TOKENIZER_DIR}/" "${MODEL_REPOSITORY}/postprocessing/config.pbtxt"
replace_config_tags '${logits_datatype}' "TYPE_FP32" "${MODEL_REPOSITORY}/postprocessing/config.pbtxt"

replace_config_tags '${triton_max_batch_size}' "128" "${MODEL_REPOSITORY}/tensorrt_llm/config.pbtxt"
replace_config_tags '${decoupled_mode}' 'true' "${MODEL_REPOSITORY}/tensorrt_llm/config.pbtxt"
replace_config_tags '${max_queue_delay_microseconds}' "1000000" "${MODEL_REPOSITORY}/tensorrt_llm/config.pbtxt"
replace_config_tags '${batching_strategy}' 'inflight_fused_batching' "${MODEL_REPOSITORY}/tensorrt_llm/config.pbtxt"
replace_config_tags '${engine_dir}' "${ENGINES_DIR}" "${MODEL_REPOSITORY}/tensorrt_llm/config.pbtxt"
replace_config_tags '${triton_backend}' "tensorrtllm" "${MODEL_REPOSITORY}/tensorrt_llm/config.pbtxt"
replace_config_tags '${max_queue_size}' "0" "${MODEL_REPOSITORY}/tensorrt_llm/config.pbtxt"
replace_config_tags '${logits_datatype}' "TYPE_FP32" "${MODEL_REPOSITORY}/tensorrt_llm/config.pbtxt"
replace_config_tags '${encoder_input_features_data_type}' "TYPE_FP32" "${MODEL_REPOSITORY}/tensorrt_llm/config.pbtxt"
}

function clone_tensorrt_llm_backend_repo {
rm -rf $TENSORRTLLM_BACKEND_DIR && mkdir $TENSORRTLLM_BACKEND_DIR
apt-get update && apt-get install git-lfs -y --no-install-recommends
git clone --single-branch --depth=1 -b ${TENSORRTLLM_BACKEND_REPO_TAG} ${TRITON_REPO_ORG}/tensorrtllm_backend.git $TENSORRTLLM_BACKEND_DIR
cd $TENSORRTLLM_BACKEND_DIR && git lfs install && git submodule update --init --recursive
}

function build_gpt2_base_model {
# Download weights from HuggingFace Transformers
cd ${GPT_DIR} && rm -rf gpt2 && git clone https://huggingface.co/gpt2-medium gpt2 && cd gpt2
rm pytorch_model.bin model.safetensors
if ! wget -q https://huggingface.co/gpt2-medium/resolve/main/pytorch_model.bin; then
echo "Downloading pytorch_model.bin failed."
exit 1
fi
cd ${GPT_DIR}

# Convert weights from HF Tranformers to FT format
python3 convert_checkpoint.py --model_dir gpt2 --dtype float16 --tp_size ${NUM_GPUS} --output_dir "./c-model/gpt2/${NUM_GPUS}-gpu/"
cd ${BASE_DIR}
}

function build_gpt2_tensorrt_engine {
# Build TensorRT engines
cd ${GPT_DIR}
trtllm-build --checkpoint_dir "./c-model/gpt2/${NUM_GPUS}-gpu/" \
--gpt_attention_plugin float16 \
--remove_input_padding enable \
--paged_kv_cache enable \
--gemm_plugin float16 \
--workers "${NUM_GPUS}" \
--output_dir "${ENGINES_DIR}"

cd ${BASE_DIR}
}
Loading