问拥抱面:期望所有的张量都在同一个设备上，但是至少发现了两个设备，cuda:0和cpu。

EN

from transformers import AutoTokenizer
from transformers import DataCollatorForSeq2Seq
from transformers import AutoModelForSeq2SeqLM, Seq2SeqTrainingArguments, Seq2SeqTrainer
from transformers import pipeline
from tqdm import tqdm
from datasets import Dataset

import pandas as pd
import numpy as np
import pyarrow as pa
import gc
import torch as t
import pickle

PATH = './datas/Batch_answers - train_data (no-blank).csv'
EPOCH = 1
LEARNING_RATE = 2e-5
TRAIN_BATCH_SIZE = 16
EVAL_BATCH_SIZE = 16
DEVICE = 'cuda' if t.cuda.is_available() else 'cpu'

df = pd.read_csv(PATH)
df = df.drop(labels='s', axis=1)
df = df.iloc[:, 1:5]
df = df.to_numpy()
qData = []

for i in tqdm(range(len(df))):
    argument = df[i][0][1:-1]
    response = df[i][1][1:-1]
    qprime = df[i][2][1:-1]
    
    qData.append({'statement':argument+'\n'+response, 'argument_sentence_summary':qprime})
    
qtable = pa.Table.from_pylist(qData)
qDataset = Dataset(qtable)
qDataset = qDataset.train_test_split(train_size=0.8)

qModel = AutoModelForSeq2SeqLM.from_pretrained("t5-small")
qTokenizer = AutoTokenizer.from_pretrained("t5-small")
qData_collator = DataCollatorForSeq2Seq(tokenizer=qTokenizer, model=qModel)

def Qpreprocessing(data):
    model_input = qTokenizer(data['statement'], max_length=250, truncation=True)
    labels = qTokenizer(text_target=data['argument_sentence_summary'], max_length=75, truncation=True)

    model_input['labels'] = labels['input_ids']
    
    return model_input

qToken = qDataset.map(Qpreprocessing, batched=True)

qTraining_args = Seq2SeqTrainingArguments(
    output_dir="./result",
    evaluation_strategy="epoch",
    learning_rate=LEARNING_RATE,
    per_device_train_batch_size=TRAIN_BATCH_SIZE,
    per_device_eval_batch_size=EVAL_BATCH_SIZE,
    weight_decay=0.01,
    save_total_limit=3,
    num_train_epochs=EPOCH,
    fp16=True,
)

qTrainer = Seq2SeqTrainer(
    model=qModel,
    args=qTraining_args,
    train_dataset=qToken['train'],
    eval_dataset=qToken['test'],
    tokenizer=qTokenizer,
    data_collator=qData_collator
)

old_collator = qTrainer.data_collator
qTrainer.data_collator = lambda data: dict(old_collator(data))
qTrainer.train()

qp = pipeline('summarization', model=qModel, tokenizer=qTokenizer)
qp(qDataset['test'][0]['statement']) #break in this line

RuntimeError                              Traceback (most recent call last)
Cell In [20], line 3
      1 qp = pipeline('summarization', model=qModel, tokenizer=qTokenizer)
      2 # temp = t.tensor(qDataset['test'][0]['statement']).to(DEVICE)
----> 3 qp(qDataset['train'][0]['statement'])

File ~\anaconda3\envs\ame\lib\site-packages\transformers\pipelines\text2text_generation.py:250, in SummarizationPipeline.__call__(self, *args, **kwargs)
    226 def __call__(self, *args, **kwargs):
    227     r"""
    228     Summarize the text(s) given as inputs.
    229 
   (...)
    248           ids of the summary.
    249     """
--> 250     return super().__call__(*args, **kwargs)

File ~\anaconda3\envs\ame\lib\site-packages\transformers\pipelines\text2text_generation.py:150, in Text2TextGenerationPipeline.__call__(self, *args, **kwargs)
    121 def __call__(self, *args, **kwargs):
    122     r"""
    123     Generate the output text(s) using text(s) given as inputs.
    124 
   (...)
    147           ids of the generated text.
    148     """
--> 150     result = super().__call__(*args, **kwargs)
    151     if (
    152         isinstance(args[0], list)
    153         and all(isinstance(el, str) for el in args[0])
    154         and all(len(res) == 1 for res in result)
    155     ):
    156         return [res[0] for res in result]

File ~\anaconda3\envs\ame\lib\site-packages\transformers\pipelines\base.py:1074, in Pipeline.__call__(self, inputs, num_workers, batch_size, *args, **kwargs)
   1072     return self.iterate(inputs, preprocess_params, forward_params, postprocess_params)
   1073 else:
-> 1074     return self.run_single(inputs, preprocess_params, forward_params, postprocess_params)

File ~\anaconda3\envs\ame\lib\site-packages\transformers\pipelines\base.py:1081, in Pipeline.run_single(self, inputs, preprocess_params, forward_params, postprocess_params)
   1079 def run_single(self, inputs, preprocess_params, forward_params, postprocess_params):
   1080     model_inputs = self.preprocess(inputs, **preprocess_params)
-> 1081     model_outputs = self.forward(model_inputs, **forward_params)
   1082     outputs = self.postprocess(model_outputs, **postprocess_params)
   1083     return outputs

File ~\anaconda3\envs\ame\lib\site-packages\transformers\pipelines\base.py:990, in Pipeline.forward(self, model_inputs, **forward_params)
    988     with inference_context():
    989         model_inputs = self._ensure_tensor_on_device(model_inputs, device=self.device)
--> 990         model_outputs = self._forward(model_inputs, **forward_params)
    991         model_outputs = self._ensure_tensor_on_device(model_outputs, device=torch.device("cpu"))
    992 else:

File ~\anaconda3\envs\ame\lib\site-packages\transformers\pipelines\text2text_generation.py:172, in Text2TextGenerationPipeline._forward(self, model_inputs, **generate_kwargs)
    170 generate_kwargs["max_length"] = generate_kwargs.get("max_length", self.model.config.max_length)
    171 self.check_inputs(input_length, generate_kwargs["min_length"], generate_kwargs["max_length"])
--> 172 output_ids = self.model.generate(**model_inputs, **generate_kwargs)
    173 out_b = output_ids.shape[0]
    174 if self.framework == "pt":

File ~\anaconda3\envs\ame\lib\site-packages\torch\autograd\grad_mode.py:27, in _DecoratorContextManager.__call__.<locals>.decorate_context(*args, **kwargs)
     24 @functools.wraps(func)
     25 def decorate_context(*args, **kwargs):
     26     with self.clone():
---> 27         return func(*args, **kwargs)

File ~\anaconda3\envs\ame\lib\site-packages\transformers\generation_utils.py:1339, in GenerationMixin.generate(self, inputs, max_length, min_length, do_sample, early_stopping, num_beams, temperature, penalty_alpha, top_k, top_p, typical_p, repetition_penalty, bad_words_ids, force_words_ids, bos_token_id, pad_token_id, eos_token_id, length_penalty, no_repeat_ngram_size, encoder_no_repeat_ngram_size, num_return_sequences, max_time, max_new_tokens, decoder_start_token_id, use_cache, num_beam_groups, diversity_penalty, prefix_allowed_tokens_fn, logits_processor, renormalize_logits, stopping_criteria, constraints, output_attentions, output_hidden_states, output_scores, return_dict_in_generate, forced_bos_token_id, forced_eos_token_id, remove_invalid_values, synced_gpus, exponential_decay_length_penalty, suppress_tokens, begin_suppress_tokens, forced_decoder_ids, **model_kwargs)
   1331         logger.warning(
   1332             "A decoder-only architecture is being used, but right-padding was detected! For correct "
   1333             "generation results, please set `padding_side='left'` when initializing the tokenizer."
   1334         )
   1336 if self.config.is_encoder_decoder and "encoder_outputs" not in model_kwargs:
   1337     # if model is encoder decoder encoder_outputs are created
   1338     # and added to `model_kwargs`
-> 1339     model_kwargs = self._prepare_encoder_decoder_kwargs_for_generation(
   1340         inputs_tensor, model_kwargs, model_input_name
   1341     )
   1343 # 4. Prepare `input_ids` which will be used for auto-regressive generation
   1344 if self.config.is_encoder_decoder:

File ~\anaconda3\envs\ame\lib\site-packages\transformers\generation_utils.py:583, in GenerationMixin._prepare_encoder_decoder_kwargs_for_generation(self, inputs_tensor, model_kwargs, model_input_name)
    581 encoder_kwargs["return_dict"] = True
    582 encoder_kwargs[model_input_name] = inputs_tensor
--> 583 model_kwargs["encoder_outputs"]: ModelOutput = encoder(**encoder_kwargs)
    585 return model_kwargs

File ~\anaconda3\envs\ame\lib\site-packages\torch\nn\modules\module.py:1130, in Module._call_impl(self, *input, **kwargs)
   1126 # If we don't have any hooks, we want to skip the rest of the logic in
   1127 # this function, and just call forward.
   1128 if not (self._backward_hooks or self._forward_hooks or self._forward_pre_hooks or _global_backward_hooks
   1129         or _global_forward_hooks or _global_forward_pre_hooks):
-> 1130     return forward_call(*input, **kwargs)
   1131 # Do not call functions when jit is used
   1132 full_backward_hooks, non_full_backward_hooks = [], []

File ~\anaconda3\envs\ame\lib\site-packages\transformers\models\t5\modeling_t5.py:941, in T5Stack.forward(self, input_ids, attention_mask, encoder_hidden_states, encoder_attention_mask, inputs_embeds, head_mask, cross_attn_head_mask, past_key_values, use_cache, output_attentions, output_hidden_states, return_dict)
    939 if inputs_embeds is None:
    940     assert self.embed_tokens is not None, "You have to initialize the model with valid token embeddings"
--> 941     inputs_embeds = self.embed_tokens(input_ids)
    943 batch_size, seq_length = input_shape
    945 # required mask seq length can be calculated via length of past

File ~\anaconda3\envs\ame\lib\site-packages\torch\nn\modules\module.py:1130, in Module._call_impl(self, *input, **kwargs)
   1126 # If we don't have any hooks, we want to skip the rest of the logic in
   1127 # this function, and just call forward.
   1128 if not (self._backward_hooks or self._forward_hooks or self._forward_pre_hooks or _global_backward_hooks
   1129         or _global_forward_hooks or _global_forward_pre_hooks):
-> 1130     return forward_call(*input, **kwargs)
   1131 # Do not call functions when jit is used
   1132 full_backward_hooks, non_full_backward_hooks = [], []

File ~\anaconda3\envs\ame\lib\site-packages\torch\nn\modules\sparse.py:158, in Embedding.forward(self, input)
    157 def forward(self, input: Tensor) -> Tensor:
--> 158     return F.embedding(
    159         input, self.weight, self.padding_idx, self.max_norm,
    160         self.norm_type, self.scale_grad_by_freq, self.sparse)

File ~\anaconda3\envs\ame\lib\site-packages\torch\nn\functional.py:2199, in embedding(input, weight, padding_idx, max_norm, norm_type, scale_grad_by_freq, sparse)
   2193     # Note [embedding_renorm set_grad_enabled]
   2194     # XXX: equivalent to
   2195     # with torch.no_grad():
   2196     #   torch.embedding_renorm_
   2197     # remove once script supports set_grad_enabled
   2198     _no_grad_embedding_renorm_(weight, input, max_norm, norm_type)
-> 2199 return torch.embedding(weight, input, padding_idx, scale_grad_by_freq, sparse)

RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu! (when checking argument for argument index in method wrapper__index_select)