How to train the model by prompting?

[liyang619]

Hi~ could I train the model and update parameters by mutate prompting code?

[infinitylogesh]

Hi, Unfortunately Mutate currently doesn't support training. However, if you wish to finetune your LLM - following the steps in the repo would probably help - https://github.com/Xirider/finetune-gpt2xl

[EricPeter]

From this from mutate import pipeline
    ...:
    ...: pipe = pipeline("text-classification-synthesis",
    ...:                 model="EleutherAI/gpt-neo-2.7B")
    ...:
    ...: task_dec ="Each item in the following contains Swahili news  texts that are classified as Kitaifa (National),Kimataifa (International), Biashara (Business),Michezo (Sports) and Burudani (Entertainment) "
    ...:
    ...:
    ...: # returns a python generator
    ...: text_synth_gen = pipe("csv",
    ...:                     data_files=["Train.csv"],
    ...:                     task_desc=task_desc,
    ...:                     text_column="content",
    ...:                     label_column="category",
    ...:                     text_column_alias="News",
    ...:                     label_column_alias="Category",
    ...:                     shot_count=5,
    ...:                     class_names=['Kimataifa', 'Burudani'])
    ...:
    ...: #Loop through the generator to synthesize examples by class
    ...: for synthesized_examples  in text_synth_gen:
    ...:     print(synthesized_examples)  

I get this error Token indices sequence length is longer than the specified maximum sequence length for this model (3341 > 2048). Running this sequence through the model will result in indexing errors
/home/ubuntu/transformers/src/transformers/generation/configuration_utils.py:422: UserWarning: num_beams is set to 1. However, early_stopping is set to True -- this flag is only used in beam-based generation modes. You should set num_beams>1 or unset early_stopping.
warnings.warn(
A decoder-only architecture is being used, but right-padding was detected! For correct generation results, please set padding_side='left' when initializing the tokenizer.
/home/ubuntu/transformers/src/transformers/generation/utils.py:1362: UserWarning: Input length of input_ids is 3341, but max_length is set to 300. This can lead to unexpected behavior. You should consider increasing max_new_tokens.
warnings.warn(

IndexError Traceback (most recent call last)
Cell In[85], line 21
10 text_synth_gen = pipe("csv",
11 data_files=["Train.csv"],
12 task_desc=task_desc,
(...)
17 shot_count=5,
18 class_names=['Kimataifa', 'Burudani'])
20 #Loop through the generator to synthesize examples by class
---> 21 for synthesized_examples in text_synth_gen:
22 print(synthesized_examples)

File ~/.local/lib/python3.8/site-packages/mutate/pipelines/text_classification.py:200, in TextClassificationSynthesize.call(self, dataset_path, text_column, label_column, task_desc, split, data_files, data_dir, text_column_alias, label_column_alias, class_names, dataset_args, dataset_kwargs, batch_size, shot_count, infinite_loop, **kwargs)
198 batch_parsed_examples = []
199 batch_class_names = []
--> 200 batch_generated_texts = self.infer.run_single_batch(
201 batch, generate_args=self.generate_kwargs
202 )
204 num_return_sequences = len(batch_generated_texts) // len(batch["input_ids"])
206 for idx, generated_text in enumerate(batch_generated_texts):

File ~/.local/lib/python3.8/site-packages/mutate/infer.py:52, in TextGeneration.run_single_batch(self, batch, is_include_prompt_in_generation, ignore_prompt_last_line, generate_args)
44 def run_single_batch(
45 self,
46 batch,
(...)
49 generate_args: Optional[Dict[str, str]] = None,
50 ):
51 batch_size = len(batch["input_ids"])
---> 52 generated_sequences = self.model.generate(
53 batch["input_ids"], attention_mask=batch["attention_mask"], **generate_args
54 )
55 generated_sequences = generated_sequences.cpu()
56 num_return_sequences = len(generated_sequences) // batch_size

File /usr/local/lib/python3.8/dist-packages/torch/utils/_contextlib.py:115, in context_decorator..decorate_context(*args, **kwargs)
112 @functools.wraps(func)
113 def decorate_context(*args, **kwargs):
114 with ctx_factory():
--> 115 return func(*args, **kwargs)

File ~/transformers/src/transformers/generation/utils.py:1763, in GenerationMixin.generate(self, inputs, generation_config, logits_processor, stopping_criteria, prefix_allowed_tokens_fn, synced_gpus, assistant_model, streamer, negative_prompt_ids, negative_prompt_attention_mask, **kwargs)
1755 input_ids, model_kwargs = self._expand_inputs_for_generation(
1756 input_ids=input_ids,
1757 expand_size=generation_config.num_return_sequences,
1758 is_encoder_decoder=self.config.is_encoder_decoder,
1759 **model_kwargs,
1760 )
1762 # 13. run sample
-> 1763 return self.sample(
1764 input_ids,
1765 logits_processor=logits_processor,
1766 logits_warper=logits_warper,
1767 stopping_criteria=stopping_criteria,
1768 pad_token_id=generation_config.pad_token_id,
1769 eos_token_id=generation_config.eos_token_id,
1770 output_scores=generation_config.output_scores,
1771 return_dict_in_generate=generation_config.return_dict_in_generate,
1772 synced_gpus=synced_gpus,
1773 streamer=streamer,
1774 **model_kwargs,
1775 )
1777 elif generation_mode == GenerationMode.BEAM_SEARCH:
1778 # 11. prepare beam search scorer
1779 beam_scorer = BeamSearchScorer(
1780 batch_size=batch_size,
1781 num_beams=generation_config.num_beams,
(...)
1786 max_length=generation_config.max_length,
1787 )

File ~/transformers/src/transformers/generation/utils.py:2860, in GenerationMixin.sample(self, input_ids, logits_processor, stopping_criteria, logits_warper, max_length, pad_token_id, eos_token_id, output_attentions, output_hidden_states, output_scores, return_dict_in_generate, synced_gpus, streamer, **model_kwargs)
2857 model_inputs = self.prepare_inputs_for_generation(input_ids, **model_kwargs)
2859 # forward pass to get next token
-> 2860 outputs = self(
2861 **model_inputs,
2862 return_dict=True,
2863 output_attentions=output_attentions,
2864 output_hidden_states=output_hidden_states,
2865 )
2867 if synced_gpus and this_peer_finished:
2868 continue # don't waste resources running the code we don't need

File /usr/local/lib/python3.8/dist-packages/torch/nn/modules/module.py:1518, in Module._wrapped_call_impl(self, *args, **kwargs)
1516 return self._compiled_call_impl(*args, **kwargs) # type: ignore[misc]
1517 else:
-> 1518 return self._call_impl(*args, **kwargs)

File /usr/local/lib/python3.8/dist-packages/torch/nn/modules/module.py:1527, in Module._call_impl(self, *args, **kwargs)
1522 # If we don't have any hooks, we want to skip the rest of the logic in
1523 # this function, and just call forward.
1524 if not (self._backward_hooks or self._backward_pre_hooks or self._forward_hooks or self._forward_pre_hooks
1525 or _global_backward_pre_hooks or _global_backward_hooks
1526 or _global_forward_hooks or _global_forward_pre_hooks):
-> 1527 return forward_call(*args, **kwargs)
1529 try:
1530 result = None

File ~/transformers/src/transformers/models/gpt_neo/modeling_gpt_neo.py:954, in GPTNeoForCausalLM.forward(self, input_ids, past_key_values, attention_mask, token_type_ids, position_ids, head_mask, inputs_embeds, labels, use_cache, output_attentions, output_hidden_states, return_dict)
946 r"""
947 labels (torch.LongTensor of shape (batch_size, sequence_length), optional):
948 Labels for language modeling. Note that the labels are shifted inside the model, i.e. you can set
949 labels = input_ids Indices are selected in [-100, 0, ..., config.vocab_size] All labels set to -100
950 are ignored (masked), the loss is only computed for labels in [0, ..., config.vocab_size]
951 """
952 return_dict = return_dict if return_dict is not None else self.config.use_return_dict
--> 954 transformer_outputs = self.transformer(
955 input_ids,
956 past_key_values=past_key_values,
957 attention_mask=attention_mask,
958 token_type_ids=token_type_ids,
959 position_ids=position_ids,
960 head_mask=head_mask,
961 inputs_embeds=inputs_embeds,
962 use_cache=use_cache,
963 output_attentions=output_attentions,
964 output_hidden_states=output_hidden_states,
965 return_dict=return_dict,
966 )
967 hidden_states = transformer_outputs[0]
969 lm_logits = self.lm_head(hidden_states)

File /usr/local/lib/python3.8/dist-packages/torch/nn/modules/module.py:1518, in Module._wrapped_call_impl(self, *args, **kwargs)
1516 return self._compiled_call_impl(*args, **kwargs) # type: ignore[misc]
1517 else:
-> 1518 return self._call_impl(*args, **kwargs)

File /usr/local/lib/python3.8/dist-packages/torch/nn/modules/module.py:1527, in Module._call_impl(self, *args, **kwargs)
1522 # If we don't have any hooks, we want to skip the rest of the logic in
1523 # this function, and just call forward.
1524 if not (self._backward_hooks or self._backward_pre_hooks or self._forward_hooks or self._forward_pre_hooks
1525 or _global_backward_pre_hooks or _global_backward_hooks
1526 or _global_forward_hooks or _global_forward_pre_hooks):
-> 1527 return forward_call(*args, **kwargs)
1529 try:
1530 result = None

File ~/transformers/src/transformers/models/gpt_neo/modeling_gpt_neo.py:778, in GPTNeoModel.forward(self, input_ids, past_key_values, attention_mask, token_type_ids, position_ids, head_mask, inputs_embeds, use_cache, output_attentions, output_hidden_states, return_dict)
776 if inputs_embeds is None:
777 inputs_embeds = self.wte(input_ids)
--> 778 position_embeds = self.wpe(position_ids)
779 hidden_states = inputs_embeds + position_embeds
781 # Attention mask.

File /usr/local/lib/python3.8/dist-packages/torch/nn/modules/module.py:1518, in Module._wrapped_call_impl(self, *args, **kwargs)
1516 return self._compiled_call_impl(*args, **kwargs) # type: ignore[misc]
1517 else:
-> 1518 return self._call_impl(*args, **kwargs)

File /usr/local/lib/python3.8/dist-packages/torch/nn/modules/module.py:1527, in Module._call_impl(self, *args, **kwargs)
1522 # If we don't have any hooks, we want to skip the rest of the logic in
1523 # this function, and just call forward.
1524 if not (self._backward_hooks or self._backward_pre_hooks or self._forward_hooks or self._forward_pre_hooks
1525 or _global_backward_pre_hooks or _global_backward_hooks
1526 or _global_forward_hooks or _global_forward_pre_hooks):
-> 1527 return forward_call(*args, **kwargs)
1529 try:
1530 result = None

File /usr/local/lib/python3.8/dist-packages/torch/nn/modules/sparse.py:162, in Embedding.forward(self, input)
161 def forward(self, input: Tensor) -> Tensor:
--> 162 return F.embedding(
163 input, self.weight, self.padding_idx, self.max_norm,
164 self.norm_type, self.scale_grad_by_freq, self.sparse)

File /usr/local/lib/python3.8/dist-packages/torch/nn/functional.py:2233, in embedding(input, weight, padding_idx, max_norm, norm_type, scale_grad_by_freq, sparse)
2227 # Note [embedding_renorm set_grad_enabled]
2228 # XXX: equivalent to
2229 # with torch.no_grad():
2230 # torch.embedding_renorm_
2231 # remove once script supports set_grad_enabled
2232 no_grad_embedding_renorm(weight, input, max_norm, norm_type)
-> 2233 return torch.embedding(weight, input, padding_idx, scale_grad_by_freq, sparse)

IndexError: index out of range in self