👑 Easy-to-use and powerful NLP and LLM library with 🤗 Awesome model zoo, supporting wide-range of NLP tasks from research to industrial applications, including 🗂Text Classification, 🔍 Neural Search, ❓ Question Answering, ℹ️ Information Extraction, 📄 Document Intelligence, 💌 Sentiment Analysis etc.
w5688414
commented
6 months ago
请提出你的问题
ValueError :( InvalidArgument ) Input ( Logits ) and Input ( Label ) should in same shape In dimensions except axis . Hint : Expected logits dime [ i ]== labels dime [ i ], but received logita dime [ i ]:56 f = labels dins [ i ]:252.]( at ../ paddle / phi / inferneta / binary . cc :909) LAUNCH INPO 2023-08-1620:46:23,137 Exit code -15 这是当时的报错，记得两次都是跑到100/3000是时候停止 GPU是四卡V100 python -m paddle.distributed.launch --gpus "0,1,2,3" finetune_generation.py \ --model_name_or_path THUDM/chatglm-6b \ --task_name_or_path ../FieldOfMachinery/ \ --max_steps 3000 \ --learning_rate 3e-5 \ --warmup_steps 20 \ --eval_steps 100 \ --logging_steps 1 \ --save_steps 1000 \ --save_total_limit 1 \ --output_dir ./checkpoints/chatglm-6b-finetune \ --src_length 64 \ --tgt_length 64 \ --per_device_eval_batch_size 4 \ --per_device_train_batch_size 4 \ --gradient_accumulation_steps 32 \ --fp16 \ --fp16_opt_level O2 \ --recompute True \ --do_train \ --do_eval \ --load_best_model_at_end True \ --tensor_parallel_degree 4 \ --do_generation True
运行单卡lora微调和单卡prefix微调时候没有问题，这个应该如何解决