Skip to content

More changes to mlperf setup to support llama2 70b and mixtral 8x7b #185

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

Merged
vipannalla merged 1 commit into from
Feb 10, 2025
Merged

More changes to mlperf setup to support llama2 70b and mixtral 8x7b #185

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
13 changes: 9 additions & 4 deletions benchmarks/mlperf/main.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23,7 +23,7 @@

import mlperf_loadgen as lg

_MLPERF_ID = "mixtral-8x7b"
_MLPERF_ID = "llama2-70b"

sys.path.insert(0, os.getcwd())

Expand Down Expand Up @@ -135,7 +135,12 @@ def get_args():
'eg. {"tok_input_len": "tok_input_length"}'
),
)

parser.add_argument(
"--mlperf-conf-id",
type=str,
default=_MLPERF_ID,
help="When given overrides the default user.conf path",
)
args = parser.parse_args()
return args

Expand All @@ -156,8 +161,8 @@ def main():
else:
user_conf = args.user_conf

settings.FromConfig(args.mlperf_conf, _MLPERF_ID, args.scenario)
settings.FromConfig(user_conf, _MLPERF_ID, args.scenario)
settings.FromConfig(args.mlperf_conf, args.mlperf_conf_id, args.scenario)
settings.FromConfig(user_conf, args.mlperf_conf_id, args.scenario)
log.info("Mlperf config: %s", args.mlperf_conf)
log.info("User config: %s", user_conf)

Expand Down
6 changes: 6 additions & 0 deletions benchmarks/mlperf/scripts/generate_server_accuracy_run.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,6 +14,9 @@

source run_utils.sh

export TOKENIZER_PATH=meta-llama/Llama-2-70b-chat-hf
export DATASET_PREFIX=""
export MODEL_ID="llama2-70b"
DATASET_NAME=$(get_dataset_name ${DATASET_TYPE})
export DATASET_PATH=${DATA_DISK_DIR}/${DATASET_NAME}.pkl
export API_URL=${API_URL}
Expand All @@ -33,6 +36,7 @@ echo "OUTPUT_ACCURACY_JSON_PATH: ${OUTPUT_ACCURACY_JSON_PATH}"
echo "USER_CONFIG: ${USER_CONFIG}"

mkdir -p ${OUTPUT_LOG_DIR} && cp ../${USER_CONFIG} ${OUTPUT_LOG_DIR}
MIXTRAL_COLS_RENAME="{\"tok_input_len\": \"tok_input_length\", \"tok_ref_output_len\": \"tok_output_length\"}"

# Accuracy Run
cd ../ && python3 main.py \
Expand All @@ -53,6 +57,8 @@ cd ../ && python3 main.py \
--tokenizer-path ${TOKENIZER_PATH} \
--log-interval ${LOG_INTERVAL} \
--num-client-threads ${NUM_CLIENT_THREADS} \
--mlperf-conf-id "${MODEL_ID}" \
--rename-dataset-cols "${MIXTRAL_COLS_RENAME}" \
--output-log-dir ${OUTPUT_LOG_DIR} 2>&1 | tee ${OUTPUT_LOG_DIR}/server_accuracy_log.log

# Eval Run
Expand Down
5 changes: 4 additions & 1 deletion benchmarks/mlperf/scripts/generate_server_audit_run.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -13,7 +13,9 @@
# limitations under the License.

source run_utils.sh

export TOKENIZER_PATH=meta-llama/Llama-2-70b-chat-hf
export DATASET_PREFIX=""
export MODEL_ID="llama2-70b"
DATASET_NAME=$(get_dataset_name ${DATASET_TYPE})
export DATASET_PATH=${DATA_DISK_DIR}/${DATASET_NAME}.pkl
export API_URL=${API_URL}
Expand Down Expand Up @@ -50,4 +52,5 @@ cd ../ && python3 main.py \
--tokenizer-path ${TOKENIZER_PATH} \
--log-interval ${LOG_INTERVAL} \
--num-client-threads ${NUM_CLIENT_THREADS} \
--mlperf-conf-id "${MODEL_ID}" \
--output-log-dir ${OUTPUT_LOG_DIR} 2>&1 | tee ${OUTPUT_LOG_DIR}/server_audit_log.log
4 changes: 4 additions & 0 deletions benchmarks/mlperf/scripts/generate_server_performance_run.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,6 +14,9 @@

source run_utils.sh

export TOKENIZER_PATH=meta-llama/Llama-2-70b-chat-hf
export DATASET_PREFIX=""
export MODEL_ID="llama2-70b"
DATASET_NAME=$(get_dataset_name ${DATASET_TYPE})
export DATASET_PATH=${DATA_DISK_DIR}/${DATASET_NAME}.pkl
export API_URL=${API_URL}
Expand Down Expand Up @@ -52,4 +55,5 @@ cd ../ && python3 main.py \
--log-interval ${LOG_INTERVAL} \
--num-client-threads ${NUM_CLIENT_THREADS} \
--rename-dataset-cols "${MIXTRAL_COLS_RENAME}" \
--mlperf-conf-id "${MODEL_ID}" \
--output-log-dir ${OUTPUT_LOG_DIR} 2>&1 | tee ${OUTPUT_LOG_DIR}/server_performance_log.log
4 changes: 2 additions & 2 deletions benchmarks/mlperf/scripts/run_utils.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15,7 +15,7 @@

# Tokenizer
# export TOKENIZER_PATH=meta-llama/Llama-2-70b-chat-hf
export DATASET_PREFIX=mixtral
export DATASET_PREFIX="mixtral-"
export TOKENIZER_PATH=mistralai/Mixtral-8x7B-Instruct-v0.1
export NUM_CLIENT_THREADS=${NUM_CLIENT_THREADS:=600}

Expand All @@ -25,7 +25,7 @@ export LOADGEN_RUN_TIMESTAMP=$(TZ=America/Los_Angeles date +%Y%m%d%H%M%S%Z)
get_dataset_name() {
dataset_type=$1
if [ ${dataset_type} = "full" ]
then echo "${DATASET_PREFIX}-processed-data"
then echo "${DATASET_PREFIX}processed-data"
elif [ ${dataset_type} = "calibration" ]
then echo "${DATASET_PREFIX}-processed-calibration-data"
fi
Expand Down
2 changes: 1 addition & 1 deletion benchmarks/mlperf/user.conf
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -22,7 +22,7 @@ mixtral-8x7b.Server.min_query_count = 15000

# These fields should be defined and overridden by user.conf.
*.Offline.target_qps = 5.0
llama2-70b.Server.target_qps = 1.0
llama2-70b.Server.target_qps = 14.5
mixtral-8x7b.Server.target_qps = 11.0


Expand Down
Loading