[X] An officially supported task in the examples folder
[X] My own task or dataset (give details below)
Reproduction
SFTTrainer showing some error like this RuntimeError: chunk expects at least a 1-dimensional tensor.
model_id = 'Qwen/Qwen2.5-32B-Instruct'
model = AutoModelForCausalLM.from_pretrained(model_id, attn_implementation="flash_attention_2", torch_dtype= torch.bfloat16, quantization_config=bnb_config,)
tokenizer = AutoTokenizer.from_pretrained(model_id, use_fast=True)
# LoRA config
peft_config = LoraConfig(
# init_lora_weights="pissa",
r=32,
lora_alpha=32,
lora_dropout=0,
bias="none",
task_type="CAUSAL_LM",
# target_modules=['up_proj', 'down_proj', 'gate_proj', 'k_proj', 'q_proj', 'v_proj', 'o_proj'],
target_modules="all-linear",
modules_to_save = ['lm_head', 'embed_token']
)
model = prepare_model_for_kbit_training(model)
def formatting_prompts_func(example):
# Extracting data from the example, checking for None values
instruction = example['instruction']
input_data = example['input_original_text'] if example['input_original_text'] is not None else ""
output_data = example['output_translation_correct_text'] if example['output_translation_correct_text'] is not None else ""
# Constructing the messages
messages = [
{'role': 'system', 'content': instruction},
{'role': 'user', 'content': input_data},
{'role': 'assistant', 'content': output_data}
]
# Applying the chat template using the provided tokenizer
texts = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=False)
# # Handle NoneType result by assigning an empty string if texts is None
# if texts is None:
# texts = ""
return {"text": texts}
# Example usage on the Hugging Face dataset
formatted_e = formatting_prompts_func(train_set[5])
print(formatted_e['text'])
# Apply the function to the entire Hugging Face dataset
train_sft = train_set.map(formatting_prompts_func)
test_sft = test_set.map(formatting_prompts_func)
HAS_BFLOAT16 = torch.cuda.is_bf16_supported()
orpo_args = SFTConfig(
# use_liger=True,
max_seq_length = 1024,
learning_rate=3e-4,
weight_decay=0.01,
warmup_steps=10,
seed=0,
lr_scheduler_type="linear",
#max_steps = 1000,
max_steps=-1,
per_device_train_batch_size=2,
per_device_eval_batch_size=2,
gradient_accumulation_steps=4,
fp16 = not HAS_BFLOAT16,
bf16 = HAS_BFLOAT16,
optim="adamw_8bit",
# optim="galore_adamw",
# optim_target_modules=optim_target_modules,
num_train_epochs=5,
eval_strategy="steps",
# remove_unused_columns=False,
# load_best_model_at_end = True,
dataset_text_field="text",
eval_steps=0.5,
save_strategy = 'steps',
logging_steps=1,
#logging_steps=10,
resume_from_checkpoint = True,
report_to="wandb",
output_dir="SFT-Qwen2.5-32b-lora-1",
# packing=True
# push_to_hub=True
)
import os
trainer = SFTTrainer(
model=model,
args=orpo_args,
train_dataset=train_sft,
eval_dataset=test_sft,
#data_collator=collator,
peft_config=peft_config,
# dataset_text_field="text",
# tokenizer=tokenizer,
processing_class=tokenizer,
# dataset_num_proc = os.cpu_count()
# dataset_num_proc=6,
)
trainer.train()
error
---------------------------------------------------------------------------
RuntimeError Traceback (most recent call last)
Cell In[23], line 1
----> 1 trainer.train()
File /usr/local/lib/python3.11/dist-packages/transformers/trainer.py:2123, in Trainer.train(self, resume_from_checkpoint, trial, ignore_keys_for_eval, **kwargs)
2121 hf_hub_utils.enable_progress_bars()
2122 else:
-> 2123 return inner_training_loop(
2124 args=args,
2125 resume_from_checkpoint=resume_from_checkpoint,
2126 trial=trial,
2127 ignore_keys_for_eval=ignore_keys_for_eval,
2128 )
File /usr/local/lib/python3.11/dist-packages/transformers/trainer.py:2481, in Trainer._inner_training_loop(self, batch_size, args, resume_from_checkpoint, trial, ignore_keys_for_eval)
2475 context = (
2476 functools.partial(self.accelerator.no_sync, model=model)
2477 if i == len(batch_samples) - 1
2478 else contextlib.nullcontext
2479 )
2480 with context():
-> 2481 tr_loss_step = self.training_step(model, inputs, num_items_in_batch)
2483 if (
2484 args.logging_nan_inf_filter
2485 and not is_torch_xla_available()
2486 and (torch.isnan(tr_loss_step) or torch.isinf(tr_loss_step))
2487 ):
2488 # if loss is nan or inf simply add the average of previous logged losses
2489 tr_loss = tr_loss + tr_loss / (1 + self.state.global_step - self._globalstep_last_logged)
File /usr/local/lib/python3.11/dist-packages/transformers/trainer.py:3579, in Trainer.training_step(self, model, inputs, num_items_in_batch)
3576 return loss_mb.reduce_mean().detach().to(self.args.device)
3578 with self.compute_loss_context_manager():
-> 3579 loss = self.compute_loss(model, inputs, num_items_in_batch=num_items_in_batch)
3581 del inputs
3582 if (
3583 self.args.torch_empty_cache_steps is not None
3584 and self.state.global_step % self.args.torch_empty_cache_steps == 0
3585 ):
File /usr/local/lib/python3.11/dist-packages/transformers/trainer.py:3633, in Trainer.compute_loss(self, model, inputs, return_outputs, num_items_in_batch)
3631 loss_kwargs["num_items_in_batch"] = num_items_in_batch
3632 inputs = {**inputs, **loss_kwargs}
-> 3633 outputs = model(**inputs)
3634 # Save past state if it exists
3635 # TODO: this needs to be fixed and made cleaner later.
3636 if self.args.past_index >= 0:
File /usr/local/lib/python3.11/dist-packages/torch/nn/modules/module.py:1553, in Module._wrapped_call_impl(self, *args, **kwargs)
1551 return self._compiled_call_impl(*args, **kwargs) # type: ignore[misc]
1552 else:
-> 1553 return self._call_impl(*args, **kwargs)
File /usr/local/lib/python3.11/dist-packages/torch/nn/modules/module.py:1562, in Module._call_impl(self, *args, **kwargs)
1557 # If we don't have any hooks, we want to skip the rest of the logic in
1558 # this function, and just call forward.
1559 if not (self._backward_hooks or self._backward_pre_hooks or self._forward_hooks or self._forward_pre_hooks
1560 or _global_backward_pre_hooks or _global_backward_hooks
1561 or _global_forward_hooks or _global_forward_pre_hooks):
-> 1562 return forward_call(*args, **kwargs)
1564 try:
1565 result = None
File /usr/local/lib/python3.11/dist-packages/torch/nn/parallel/data_parallel.py:176, in DataParallel.forward(self, *inputs, **kwargs)
171 if t.device != self.src_device_obj:
172 raise RuntimeError("module must have its parameters and buffers "
173 f"on device {self.src_device_obj} (device_ids[0]) but found one of "
174 f"them on device: {t.device}")
--> 176 inputs, module_kwargs = self.scatter(inputs, kwargs, self.device_ids)
177 # for forward function without any inputs, empty list and dict will be created
178 # so the module can be executed on one device which is the first one in device_ids
179 if not inputs and not module_kwargs:
File /usr/local/lib/python3.11/dist-packages/torch/nn/parallel/data_parallel.py:198, in DataParallel.scatter(self, inputs, kwargs, device_ids)
192 def scatter(
193 self,
194 inputs: Tuple[Any, ...],
195 kwargs: Optional[Dict[str, Any]],
196 device_ids: Sequence[Union[int, torch.device]],
197 ) -> Any:
--> 198 return scatter_kwargs(inputs, kwargs, device_ids, dim=self.dim)
File /usr/local/lib/python3.11/dist-packages/torch/nn/parallel/scatter_gather.py:78, in scatter_kwargs(inputs, kwargs, target_gpus, dim)
76 r"""Scatter with support for kwargs dictionary."""
77 scattered_inputs = scatter(inputs, target_gpus, dim) if inputs else []
---> 78 scattered_kwargs = scatter(kwargs, target_gpus, dim) if kwargs else []
79 if len(scattered_inputs) < len(scattered_kwargs):
80 scattered_inputs.extend(() for _ in range(len(scattered_kwargs) - len(scattered_inputs)))
File /usr/local/lib/python3.11/dist-packages/torch/nn/parallel/scatter_gather.py:64, in scatter(inputs, target_gpus, dim)
58 # After scatter_map is called, a scatter_map cell will exist. This cell
59 # has a reference to the actual function scatter_map, which has references
60 # to a closure that has a reference to the scatter_map cell (because the
61 # fn is recursive). To avoid this reference cycle, we set the function to
62 # None, clearing the cell
63 try:
---> 64 res = scatter_map(inputs)
65 finally:
66 scatter_map = None # type: ignore[assignment]
File /usr/local/lib/python3.11/dist-packages/torch/nn/parallel/scatter_gather.py:55, in scatter.<locals>.scatter_map(obj)
53 return [list(i) for i in zip(*map(scatter_map, obj))]
54 if isinstance(obj, dict) and len(obj) > 0:
---> 55 return [type(obj)(i) for i in zip(*map(scatter_map, obj.items()))]
56 return [obj for _ in target_gpus]
File /usr/local/lib/python3.11/dist-packages/torch/nn/parallel/scatter_gather.py:51, in scatter.<locals>.scatter_map(obj)
49 return [type(obj)(*args) for args in zip(*map(scatter_map, obj))]
50 if isinstance(obj, tuple) and len(obj) > 0:
---> 51 return list(zip(*map(scatter_map, obj)))
52 if isinstance(obj, list) and len(obj) > 0:
53 return [list(i) for i in zip(*map(scatter_map, obj))]
File /usr/local/lib/python3.11/dist-packages/torch/nn/parallel/scatter_gather.py:47, in scatter.<locals>.scatter_map(obj)
45 def scatter_map(obj):
46 if isinstance(obj, torch.Tensor):
---> 47 return Scatter.apply(target_gpus, None, dim, obj)
48 if _is_namedtuple(obj):
49 return [type(obj)(*args) for args in zip(*map(scatter_map, obj))]
File /usr/local/lib/python3.11/dist-packages/torch/autograd/function.py:574, in Function.apply(cls, *args, **kwargs)
571 if not torch._C._are_functorch_transforms_active():
572 # See NOTE: [functorch vjp and autograd interaction]
573 args = _functorch.utils.unwrap_dead_wrappers(args)
--> 574 return super().apply(*args, **kwargs) # type: ignore[misc]
576 if not is_setup_ctx_defined:
577 raise RuntimeError(
578 "In order to use an autograd.Function with functorch transforms "
579 "(vmap, grad, jvp, jacrev, ...), it must override the setup_context "
580 "staticmethod. For more details, please see "
581 "https://pytorch.org/docs/main/notes/extending.func.html"
582 )
File /usr/local/lib/python3.11/dist-packages/torch/nn/parallel/_functions.py:96, in Scatter.forward(ctx, target_gpus, chunk_sizes, dim, input)
93 if torch.cuda.is_available() and ctx.input_device == -1:
94 # Perform CPU to GPU copies in a background stream
95 streams = [_get_stream(torch.device("cuda", device)) for device in target_gpus]
---> 96 outputs = comm.scatter(input, target_gpus, chunk_sizes, ctx.dim, streams)
97 # Synchronize with the copy stream
98 if streams is not None:
File /usr/local/lib/python3.11/dist-packages/torch/nn/parallel/comm.py:188, in scatter(tensor, devices, chunk_sizes, dim, streams, out)
186 if out is None:
187 devices = [_get_device_index(d) for d in devices]
--> 188 return tuple(torch._C._scatter(tensor, devices, chunk_sizes, dim, streams))
189 else:
190 if devices is not None:
RuntimeError: chunk expects at least a 1-dimensional tensor
Expected behavior
RuntimeError: chunk expects at least a 1-dimensional tensor
imrankh46
commented
3 hours ago
System Info
Name: trl Version: 0.13.0.dev0 Name: transformers Version: 4.46.2 Python 3.11.10
Information
Tasks
examplesfolderReproduction
SFTTrainer showing some error like this RuntimeError: chunk expects at least a 1-dimensional tensor.
error
Expected behavior
RuntimeError: chunk expects at least a 1-dimensional tensor