Skip to content

[PaddleNLP]Fix llm for LLama-Factory #1817

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
Liujie0926 wants to merge 3 commits into
base: develop
Choose a base branch
from
Open

[PaddleNLP]Fix llm for LLama-Factory #1817

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
f1befd7
fix docker_images.yaml
Liujie0926 Feb 6, 2025
35a9c41
Merge branch 'PaddlePaddle:develop' into develop
Liujie0926 Mar 17, 2025
4130764
update LLama-Factory
Liujie0926 Mar 17, 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
3 changes: 2 additions & 1 deletion frame_benchmark/pytorch/dynamic/PaddleNLP/scripts/llama2/benchmark_common/analysis_log.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -297,7 +297,8 @@ def get_loss(self):
line_words = line.split(self.separator) if self.separator else line.split()
for i in range(len(line_words) - 1):
if line_words[i] == self.convergence_key:
result_loss = line_words[i + 1]
# train_tokens_per_second = 11210.536
result_loss = line_words[i + 2]
result_loss = result_loss.replace(',', '')
raise ExceptionTest()
except ExceptionTest:
Expand Down
81 changes: 44 additions & 37 deletions frame_benchmark/pytorch/dynamic/PaddleNLP/scripts/llama2/benchmark_common/prepare.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -25,6 +25,7 @@ mkdir -p /opt/${model_name_or_path} && cd /opt/${model_name_or_path}
export no_proxy=bcebos.com
case ${model_name_or_path} in
shakechen/Llama-2-7b-hf)
echo '
wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/shakechen/Llama-2-7b-hf/LICENSE.txt
wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/shakechen/Llama-2-7b-hf/README.md
wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/shakechen/Llama-2-7b-hf/Responsible-Use-Guide.pdf
Expand All @@ -42,8 +43,11 @@ shakechen/Llama-2-7b-hf)
wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/shakechen/Llama-2-7b-hf/tokenizer.json
wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/shakechen/Llama-2-7b-hf/tokenizer.model
wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/shakechen/Llama-2-7b-hf/tokenizer_config.json
' > wget_7b.sh
bash wget_7b.sh
echo "download models for shakechen/Llama-2-7b-hf done" ;;
ydyajyA/Llama-2-13b-chat-hf)
echo '
wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/ydyajyA/Llama-2-13b-chat-hf/config.json
wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/ydyajyA/Llama-2-13b-chat-hf/configuration.json
wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/ydyajyA/Llama-2-13b-chat-hf/generation_config.json
Expand All @@ -60,27 +64,29 @@ ydyajyA/Llama-2-13b-chat-hf)
wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/ydyajyA/Llama-2-13b-chat-hf/tokenizer.model
wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/ydyajyA/Llama-2-13b-chat-hf/tokenizer_config.json
wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/ydyajyA/Llama-2-13b-chat-hf/up.ipynb
' > wget_13b.sh
bash wget_13b.sh
echo "download models for ydyajyA/Llama-2-13b-chat-hf done" ;;
meta-llama/Llama-2-70b-hf)
nums1=("00001" "00002" "00003" "00004" "00005")
for num in "${nums1[@]}"; do
url="https://paddlenlp.bj.bcebos.com/models/huggingface/meta-llama/Llama-2-70b-hf/model-${num}-of-00015.safetensors"
wget -c "${url}" &
done
wait
nums2=("00006" "00007" "00008" "00009" "00010")
for num in "${nums2[@]}"; do
url="https://paddlenlp.bj.bcebos.com/models/huggingface/meta-llama/Llama-2-70b-hf/model-${num}-of-00015.safetensors"
wget -c "${url}" &
done
wait
nums3=("00011" "00012" "00013" "00014" "00015")
for num in "${nums3[@]}"; do
url="https://paddlenlp.bj.bcebos.com/models/huggingface/meta-llama/Llama-2-70b-hf/model-${num}-of-00015.safetensors"
wget -c "${url}" &
done
wait

# nums1=("00001" "00002" "00003" "00004" "00005")
# for num in "${nums1[@]}"; do
# url="https://paddlenlp.bj.bcebos.com/models/huggingface/meta-llama/Llama-2-70b-hf/model-${num}-of-00015.safetensors"
# wget -c "${url}" &
# done
# wait
# nums2=("00006" "00007" "00008" "00009" "00010")
# for num in "${nums2[@]}"; do
# url="https://paddlenlp.bj.bcebos.com/models/huggingface/meta-llama/Llama-2-70b-hf/model-${num}-of-00015.safetensors"
# wget -c "${url}" &
# done
# wait
# nums3=("00011" "00012" "00013" "00014" "00015")
# for num in "${nums3[@]}"; do
# url="https://paddlenlp.bj.bcebos.com/models/huggingface/meta-llama/Llama-2-70b-hf/model-${num}-of-00015.safetensors"
# wget -c "${url}" &
# done
# wait
echo '
wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/meta-llama/Llama-2-70b-hf/LICENSE.txt
wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/meta-llama/Llama-2-70b-hf/README.md
wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/meta-llama/Llama-2-70b-hf/Responsible-Use-Guide.pdf
Expand All @@ -89,25 +95,27 @@ meta-llama/Llama-2-70b-hf)
wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/meta-llama/Llama-2-70b-hf/configuration.json
wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/meta-llama/Llama-2-70b-hf/generation_config.json
wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/meta-llama/Llama-2-70b-hf/llama_updates.patch
# wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/meta-llama/Llama-2-70b-hf/model-00001-of-00015.safetensors
# wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/meta-llama/Llama-2-70b-hf/model-00002-of-00015.safetensors
# wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/meta-llama/Llama-2-70b-hf/model-00003-of-00015.safetensors
# wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/meta-llama/Llama-2-70b-hf/model-00004-of-00015.safetensors
# wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/meta-llama/Llama-2-70b-hf/model-00005-of-00015.safetensors
# wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/meta-llama/Llama-2-70b-hf/model-00006-of-00015.safetensors
# wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/meta-llama/Llama-2-70b-hf/model-00007-of-00015.safetensors
# wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/meta-llama/Llama-2-70b-hf/model-00008-of-00015.safetensors
# wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/meta-llama/Llama-2-70b-hf/model-00009-of-00015.safetensors
# wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/meta-llama/Llama-2-70b-hf/model-00010-of-00015.safetensors
# wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/meta-llama/Llama-2-70b-hf/model-00011-of-00015.safetensors
# wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/meta-llama/Llama-2-70b-hf/model-00012-of-00015.safetensors
# wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/meta-llama/Llama-2-70b-hf/model-00013-of-00015.safetensors
# wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/meta-llama/Llama-2-70b-hf/model-00014-of-00015.safetensors
# wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/meta-llama/Llama-2-70b-hf/model-00015-of-00015.safetensors
wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/meta-llama/Llama-2-70b-hf/model-00001-of-00015.safetensors
wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/meta-llama/Llama-2-70b-hf/model-00002-of-00015.safetensors
wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/meta-llama/Llama-2-70b-hf/model-00003-of-00015.safetensors
wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/meta-llama/Llama-2-70b-hf/model-00004-of-00015.safetensors
wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/meta-llama/Llama-2-70b-hf/model-00005-of-00015.safetensors
wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/meta-llama/Llama-2-70b-hf/model-00006-of-00015.safetensors
wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/meta-llama/Llama-2-70b-hf/model-00007-of-00015.safetensors
wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/meta-llama/Llama-2-70b-hf/model-00008-of-00015.safetensors
wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/meta-llama/Llama-2-70b-hf/model-00009-of-00015.safetensors
wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/meta-llama/Llama-2-70b-hf/model-00010-of-00015.safetensors
wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/meta-llama/Llama-2-70b-hf/model-00011-of-00015.safetensors
wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/meta-llama/Llama-2-70b-hf/model-00012-of-00015.safetensors
wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/meta-llama/Llama-2-70b-hf/model-00013-of-00015.safetensors
wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/meta-llama/Llama-2-70b-hf/model-00014-of-00015.safetensors
wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/meta-llama/Llama-2-70b-hf/model-00015-of-00015.safetensors
wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/meta-llama/Llama-2-70b-hf/model.safetensors.index.json
wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/meta-llama/Llama-2-70b-hf/special_tokens_map.json
wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/meta-llama/Llama-2-70b-hf/tokenizer.model
wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/meta-llama/Llama-2-70b-hf/tokenizer_config.json
' > wget_70b.sh
bash wget_70b.sh
echo "download models for meta-llama/Llama-2-70b-hf done" ;;
*)
echo "${model_name_or_path} not in bos, download from modelscope";
Expand All @@ -119,8 +127,7 @@ esac
cd -


mv -v data data_bak
wget --no-proxy -c https://paddlenlp.bj.bcebos.com/llm_benchmark_data/llamafactory_data.tar.gz
tar zxf llamafactory_data.tar.gz && rm -rf llamafactory_data.tar.gz
tar zxf llamafactory_data.tar.gz
wget --no-proxy -c https://paddlenlp.bj.bcebos.com/llm_benchmark_data/deepspeed.tar.gz
tar zxf deepspeed.tar.gz && rm -rf deepspeed.tar.gz
tar zxf deepspeed.tar.gz
12 changes: 6 additions & 6 deletions frame_benchmark/pytorch/dynamic/PaddleNLP/scripts/llama2/benchmark_common/run_benchmark.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15,7 +15,7 @@ function _set_params(){
speed_unit="effective_tokens/sec" # (必选)速度指标单位
skip_steps=0 # (必选)解析日志，跳过模型前几个性能不稳定的step
keyword="effective_tokens_per_sec" # (必选)解析日志，筛选出性能数据所在行的关键字
convergence_key="total_tokens:" # (可选)解析日志，筛选出收敛数据所在行的关键字 如：convergence_key="loss:"
convergence_key="train_tokens_per_second" # (可选)解析日志，筛选出收敛数据所在行的关键字 如：convergence_key="loss:"
max_iter=${8:-"100"} # （可选）需保证模型执行时间在5分钟内，需要修改代码提前中断的直接提PR 合入套件 或是max_epoch
num_workers=${9:-"3"} # (可选)
is_large_model=True
Expand Down Expand Up @@ -91,11 +91,11 @@ function _train(){
export TOKENS_PER_STEP=131072
fi
timeout 40m ${train_cmd} > ${log_file} 2>&1
Tokens_per_second_per_gpu=`cat ${log_file} | grep 'train_samples_per_second =' \
|awk -F'= ' '{print $2}' |awk -F' ' '{print $1}'`
length=4096
Total_Tokens_per_second_per_gpu=$(awk -v a="$Tokens_per_second_per_gpu" -v b="$length" 'BEGIN {printf "%.2f\n", a * b}')
echo "total_tokens: ${Total_Tokens_per_second_per_gpu}" >> ${log_file}
# Tokens_per_second_per_gpu=`cat ${log_file} | grep 'train_samples_per_second =' \
# |awk -F'= ' '{print $2}' |awk -F' ' '{print $1}'`
# length=4096
# Total_Tokens_per_second_per_gpu=$(awk -v a="$Tokens_per_second_per_gpu" -v b="$length" 'BEGIN {printf "%.2f\n", a * b}')
# echo "total_tokens: ${Total_Tokens_per_second_per_gpu}" >> ${log_file}
# 这个判断，无论是否成功都是0
if [ $? -ne 0 ];then
echo -e "${model_name}, FAIL"
Expand Down
3 changes: 2 additions & 1 deletion frame_benchmark/pytorch/dynamic/PaddleNLP/scripts/qwen2_5/benchmark_common/analysis_log.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -297,7 +297,8 @@ def get_loss(self):
line_words = line.split(self.separator) if self.separator else line.split()
for i in range(len(line_words) - 1):
if line_words[i] == self.convergence_key:
result_loss = line_words[i + 1]
# train_tokens_per_second = 11210.536
result_loss = line_words[i + 2]
result_loss = result_loss.replace(',', '')
raise ExceptionTest()
except ExceptionTest:
Expand Down
Loading