github
/
ColossalAI
mirror of https://github.com/hpcaitech/ColossalAI
1
0
Fork 0
Code Issues Projects Releases Wiki Activity

[hotfix] fix gpt gemini example (#2404) 

* [hotfix] fix gpt gemini example

* [example] add new assertions
pull/2405/head
HELSON 2023-01-09 15:52:17 +08:00 committed by GitHub
parent 9880fd2cd8
commit 498b5ca993
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
2 changed files with 18 additions and 14 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

30
examples/language/gpt/gemini/benchmark_gemini.sh
View File

@ -1,18 +1,20 @@
for MODEL_TYPE in "gpt2_medium"; do for MODEL_TYPE in "gpt2_medium"; do
for BATCH_SIZE in 16; do for DISPAN in "colossalai"; do
for GPUNUM in 1 2 4 8; do for BATCH_SIZE in 16; do
for TPDEGREE in 1 2 4 8; do for GPUNUM in 1 2 4 8; do
if [ ${TPDEGREE} -gt ${GPUNUM} ]; then for TPDEGREE in 1 2 4 8; do
continue if [ ${TPDEGREE} -gt ${GPUNUM} ]; then
fi continue
for PLACEMENT in "cpu" "auto"; do fi
echo "****************** Begin ***************************" for PLACEMENT in "cpu" "auto"; do
echo "* benchmrking MODEL_TYPE ${MODEL_TYPE} BS ${BATCH_SIZE} BS ${BS} GPUNUM ${GPUNUM} TPDEGREE ${TPDEGREE} PLACEMENT ${PLACEMENT}" echo "****************** Begin ***************************"
MODEL_TYPE=${MODEL_TYPE} BATCH_SIZE=${BATCH_SIZE} GPUNUM=${GPUNUM} TPDEGREE=${TPDEGREE} PLACEMENT=${PLACEMENT} \ echo "+ benchmrking MODEL ${MODEL_TYPE} DISPAN ${DISPAN} GPU ${GPUNUM} BS ${BATCH_SIZE} TP ${TPDEGREE} POLICY ${PLACEMENT}"
bash ./gemini/run_gemini.sh MODEL_TYPE=${MODEL_TYPE} DISPAN=${DISPAN} BATCH_SIZE=${BATCH_SIZE} GPUNUM=${GPUNUM} TPDEGREE=${TPDEGREE} PLACEMENT=${PLACEMENT} \
echo "****************** Finished ***************************" bash ./run_gemini.sh
echo "" echo "****************** Finished ***************************"
echo "" echo ""
echo ""
done
done done
done done
done done

2
examples/language/gpt/gemini/train_gpt_demo.py
View File

@ -270,6 +270,7 @@ def main():
tp_pg = ProcessGroup(tp_degree=args.tp_degree) tp_pg = ProcessGroup(tp_degree=args.tp_degree)
# Tensor Parallelism (TP) # Tensor Parallelism (TP)
# You should notice that v0.1.10 is not compatible with TP degree > 1
tensor_parallelize(model, tp_pg) tensor_parallelize(model, tp_pg)
# build a Gemini model and a highly optimized cpu optimizer # build a Gemini model and a highly optimized cpu optimizer
@ -278,6 +279,7 @@ def main():
logger.info(get_mem_info(prefix='After init optim, '), ranks=[0]) logger.info(get_mem_info(prefix='After init optim, '), ranks=[0])
else: else:
assert args.tp_degree == 1, "The degree of TP should be 1 for DDP examples."
model = model_builder(args.model_type)(checkpoint=True).cuda() model = model_builder(args.model_type)(checkpoint=True).cuda()
if args.distplan.startswith("torch"): if args.distplan.startswith("torch"):