diff --git a/frame_benchmark/pytorch/dynamic/PaddleNLP/scripts/llama2/benchmark_common/analysis_log.py b/frame_benchmark/pytorch/dynamic/PaddleNLP/scripts/llama2/benchmark_common/analysis_log.py index 7ff99b6a00..e393c27f1a 100644 --- a/frame_benchmark/pytorch/dynamic/PaddleNLP/scripts/llama2/benchmark_common/analysis_log.py +++ b/frame_benchmark/pytorch/dynamic/PaddleNLP/scripts/llama2/benchmark_common/analysis_log.py @@ -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: diff --git a/frame_benchmark/pytorch/dynamic/PaddleNLP/scripts/llama2/benchmark_common/prepare.sh b/frame_benchmark/pytorch/dynamic/PaddleNLP/scripts/llama2/benchmark_common/prepare.sh index 72f47b3212..58931e4c6a 100644 --- a/frame_benchmark/pytorch/dynamic/PaddleNLP/scripts/llama2/benchmark_common/prepare.sh +++ b/frame_benchmark/pytorch/dynamic/PaddleNLP/scripts/llama2/benchmark_common/prepare.sh @@ -25,6 +25,7 @@ mkdir -p /opt/${model_name_or_path} && cd /opt/${model_name_or_path} export no_proxy=bcebos.com case ${model_name_or_path} in shakechen/Llama-2-7b-hf) + echo ' wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/shakechen/Llama-2-7b-hf/LICENSE.txt wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/shakechen/Llama-2-7b-hf/README.md wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/shakechen/Llama-2-7b-hf/Responsible-Use-Guide.pdf @@ -42,8 +43,11 @@ shakechen/Llama-2-7b-hf) wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/shakechen/Llama-2-7b-hf/tokenizer.json wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/shakechen/Llama-2-7b-hf/tokenizer.model wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/shakechen/Llama-2-7b-hf/tokenizer_config.json + ' > wget_7b.sh + bash wget_7b.sh echo "download models for shakechen/Llama-2-7b-hf done" ;; ydyajyA/Llama-2-13b-chat-hf) + echo ' wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/ydyajyA/Llama-2-13b-chat-hf/config.json wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/ydyajyA/Llama-2-13b-chat-hf/configuration.json wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/ydyajyA/Llama-2-13b-chat-hf/generation_config.json @@ -60,27 +64,29 @@ ydyajyA/Llama-2-13b-chat-hf) wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/ydyajyA/Llama-2-13b-chat-hf/tokenizer.model wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/ydyajyA/Llama-2-13b-chat-hf/tokenizer_config.json wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/ydyajyA/Llama-2-13b-chat-hf/up.ipynb + ' > wget_13b.sh + bash wget_13b.sh echo "download models for ydyajyA/Llama-2-13b-chat-hf done" ;; meta-llama/Llama-2-70b-hf) - nums1=("00001" "00002" "00003" "00004" "00005") - for num in "${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 @@ -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"; @@ -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 diff --git a/frame_benchmark/pytorch/dynamic/PaddleNLP/scripts/llama2/benchmark_common/run_benchmark.sh b/frame_benchmark/pytorch/dynamic/PaddleNLP/scripts/llama2/benchmark_common/run_benchmark.sh index 4b44aca87d..47b1b19a67 100644 --- a/frame_benchmark/pytorch/dynamic/PaddleNLP/scripts/llama2/benchmark_common/run_benchmark.sh +++ b/frame_benchmark/pytorch/dynamic/PaddleNLP/scripts/llama2/benchmark_common/run_benchmark.sh @@ -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 @@ -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" diff --git a/frame_benchmark/pytorch/dynamic/PaddleNLP/scripts/qwen2_5/benchmark_common/analysis_log.py b/frame_benchmark/pytorch/dynamic/PaddleNLP/scripts/qwen2_5/benchmark_common/analysis_log.py index 7ff99b6a00..e393c27f1a 100644 --- a/frame_benchmark/pytorch/dynamic/PaddleNLP/scripts/qwen2_5/benchmark_common/analysis_log.py +++ b/frame_benchmark/pytorch/dynamic/PaddleNLP/scripts/qwen2_5/benchmark_common/analysis_log.py @@ -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: diff --git a/frame_benchmark/pytorch/dynamic/PaddleNLP/scripts/qwen2_5/benchmark_common/prepare.sh b/frame_benchmark/pytorch/dynamic/PaddleNLP/scripts/qwen2_5/benchmark_common/prepare.sh index d260a274d1..9eb3b757e5 100644 --- a/frame_benchmark/pytorch/dynamic/PaddleNLP/scripts/qwen2_5/benchmark_common/prepare.sh +++ b/frame_benchmark/pytorch/dynamic/PaddleNLP/scripts/qwen2_5/benchmark_common/prepare.sh @@ -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 Qwen/Qwen2.5-1.5B) + echo ' wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-1.5B/LICENSE wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-1.5B/README.md wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-1.5B/config.json @@ -35,8 +36,11 @@ Qwen/Qwen2.5-1.5B) wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-1.5B/tokenizer.json wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-1.5B/tokenizer_config.json wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-1.5B/vocab.json + ' > wget_1_5b.sh + bash wget_1_5b.sh echo "download models for Qwen/Qwen2.5-1.5B done" ;; Qwen/Qwen2.5-7B) + echo ' wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-7B/LICENSE wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-7B/README.md wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-7B/config.json @@ -51,8 +55,11 @@ Qwen/Qwen2.5-7B) wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-7B/tokenizer.json wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-7B/tokenizer_config.json wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-7B/vocab.json + ' > wget_7b.sh + bash wget_7b.sh echo "download models for Qwen/Qwen2.5-7B done" ;; Qwen/Qwen2.5-14B) + echo ' wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-14B/LICENSE wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-14B/README.md wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-14B/config.json @@ -78,92 +85,96 @@ Qwen/Qwen2.5-14B) wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-14B/tokenizer.json wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-14B/tokenizer_config.json wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-14B/vocab.json + ' > wget_14b.sh + bash wget_14b.sh echo "download models for Qwen/Qwen2.5-14B done" ;; Qwen/Qwen2.5-72B) - nums1=("00001" "00002" "00003" "00004" "00005" "00006" "00007") - for num in "${nums1[@]}"; do - url="https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-${num}-of-00037.safetensors" - wget -c "${url}" & - done - wait - nums2=("00008" "00009" "00010" "00011" "00012" "00013") - for num in "${nums2[@]}"; do - url="https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-${num}-of-00037.safetensors" - wget -c "${url}" & - done - wait - nums3=("00014" "00015" "00016" "00017" "00018" "00019") - for num in "${nums3[@]}"; do - url="https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-${num}-of-00037.safetensors" - wget -c "${url}" & - done - wait - nums4=("00020" "00021" "00022" "00023" "00024" "00025") - for num in "${nums4[@]}"; do - url="https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-${num}-of-00037.safetensors" - wget -c "${url}" & - done - wait - nums5=("00026" "00027" "00028" "00029" "00030" "00031") - for num in "${nums5[@]}"; do - url="https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-${num}-of-00037.safetensors" - wget -c "${url}" & - done - wait - nums6=("00032" "00033" "00034" "00035" "00036" "00037") - for num in "${nums6[@]}"; do - url="https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-${num}-of-00037.safetensors" - wget -c "${url}" & - done - wait - + # nums1=("00001" "00002" "00003" "00004" "00005" "00006" "00007") + # for num in "${nums1[@]}"; do + # url="https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-${num}-of-00037.safetensors" + # wget -c "${url}" & + # done + # wait + # nums2=("00008" "00009" "00010" "00011" "00012" "00013") + # for num in "${nums2[@]}"; do + # url="https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-${num}-of-00037.safetensors" + # wget -c "${url}" & + # done + # wait + # nums3=("00014" "00015" "00016" "00017" "00018" "00019") + # for num in "${nums3[@]}"; do + # url="https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-${num}-of-00037.safetensors" + # wget -c "${url}" & + # done + # wait + # nums4=("00020" "00021" "00022" "00023" "00024" "00025") + # for num in "${nums4[@]}"; do + # url="https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-${num}-of-00037.safetensors" + # wget -c "${url}" & + # done + # wait + # nums5=("00026" "00027" "00028" "00029" "00030" "00031") + # for num in "${nums5[@]}"; do + # url="https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-${num}-of-00037.safetensors" + # wget -c "${url}" & + # done + # wait + # nums6=("00032" "00033" "00034" "00035" "00036" "00037") + # for num in "${nums6[@]}"; do + # url="https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-${num}-of-00037.safetensors" + # wget -c "${url}" & + # done + # wait + echo ' wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/LICENSE wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/README.md wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/config.json wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/configuration.json wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/generation_config.json wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/merges.txt - # wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00001-of-00037.safetensors - # wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00002-of-00037.safetensors - # wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00003-of-00037.safetensors - # wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00004-of-00037.safetensors - # wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00005-of-00037.safetensors - # wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00006-of-00037.safetensors - # wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00007-of-00037.safetensors - # wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00008-of-00037.safetensors - # wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00009-of-00037.safetensors - # wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00010-of-00037.safetensors - # wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00011-of-00037.safetensors - # wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00012-of-00037.safetensors - # wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00013-of-00037.safetensors - # wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00014-of-00037.safetensors - # wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00015-of-00037.safetensors - # wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00016-of-00037.safetensors - # wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00017-of-00037.safetensors - # wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00018-of-00037.safetensors - # wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00019-of-00037.safetensors - # wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00020-of-00037.safetensors - # wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00021-of-00037.safetensors - # wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00022-of-00037.safetensors - # wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00023-of-00037.safetensors - # wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00024-of-00037.safetensors - # wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00025-of-00037.safetensors - # wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00026-of-00037.safetensors - # wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00027-of-00037.safetensors - # wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00028-of-00037.safetensors - # wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00029-of-00037.safetensors - # wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00030-of-00037.safetensors - # wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00031-of-00037.safetensors - # wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00032-of-00037.safetensors - # wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00033-of-00037.safetensors - # wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00034-of-00037.safetensors - # wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00035-of-00037.safetensors - # wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00036-of-00037.safetensors - # wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00037-of-00037.safetensors + wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00001-of-00037.safetensors + wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00002-of-00037.safetensors + wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00003-of-00037.safetensors + wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00004-of-00037.safetensors + wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00005-of-00037.safetensors + wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00006-of-00037.safetensors + wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00007-of-00037.safetensors + wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00008-of-00037.safetensors + wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00009-of-00037.safetensors + wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00010-of-00037.safetensors + wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00011-of-00037.safetensors + wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00012-of-00037.safetensors + wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00013-of-00037.safetensors + wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00014-of-00037.safetensors + wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00015-of-00037.safetensors + wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00016-of-00037.safetensors + wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00017-of-00037.safetensors + wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00018-of-00037.safetensors + wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00019-of-00037.safetensors + wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00020-of-00037.safetensors + wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00021-of-00037.safetensors + wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00022-of-00037.safetensors + wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00023-of-00037.safetensors + wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00024-of-00037.safetensors + wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00025-of-00037.safetensors + wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00026-of-00037.safetensors + wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00027-of-00037.safetensors + wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00028-of-00037.safetensors + wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00029-of-00037.safetensors + wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00030-of-00037.safetensors + wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00031-of-00037.safetensors + wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00032-of-00037.safetensors + wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00033-of-00037.safetensors + wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00034-of-00037.safetensors + wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00035-of-00037.safetensors + wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00036-of-00037.safetensors + wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model-00037-of-00037.safetensors wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/model.safetensors.index.json wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/tokenizer.json wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/tokenizer_config.json wget --no-proxy -c https://paddlenlp.bj.bcebos.com/models/huggingface/Qwen/Qwen2.5-72B/vocab.json + ' > wget_72b.sh + bash wget_72b.sh echo "download models for Qwen/Qwen2.5-72B done" ;; *) echo "${model_name_or_path} not in bos, download from modelscope"; @@ -174,9 +185,7 @@ Qwen/Qwen2.5-72B) 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 diff --git a/frame_benchmark/pytorch/dynamic/PaddleNLP/scripts/qwen2_5/benchmark_common/run_benchmark.sh b/frame_benchmark/pytorch/dynamic/PaddleNLP/scripts/qwen2_5/benchmark_common/run_benchmark.sh index 4b44aca87d..47b1b19a67 100644 --- a/frame_benchmark/pytorch/dynamic/PaddleNLP/scripts/qwen2_5/benchmark_common/run_benchmark.sh +++ b/frame_benchmark/pytorch/dynamic/PaddleNLP/scripts/qwen2_5/benchmark_common/run_benchmark.sh @@ -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 @@ -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"