enable pretrain.

[fix] fix genarate with custom models does not go to custom_models

.vscode/launch.json

@@ -4,25 +4,15 @@
     // 欲了解更多信息，请访问: https://go.microsoft.com/fwlink/?linkid=830387
     "version": "0.2.0",
     "configurations": [
-        {
-            "name": "Python: train",
-            "type": "python",
-            "request": "launch",
-            "program": "train.py",
-            "args": [
-                "--dataset_name", "wikitext:wikitext-2-v1",
-            ],
-            "console": "integratedTerminal",
-            "justMyCode": true
-        },
         {
             "name": "Python: generate",
-            "type": "python",
+            "type": "debugpy",
             "request": "launch",
-            "program": "generate.py",
+            "program": "${file}",
             "args": [],
+            "cwd": "${fileDirname}",
             "console": "integratedTerminal",
-            "justMyCode": true
+            "justMyCode": false
         }
     ]
 }

README.md

@@ -1,9 +1,51 @@
 # GPT-Pretrain
 
-## Usage
+# Usage
+
+## Make it simple
 
 ```
-python lit_train.py --model_name gpt2
+python lit_train.py --model_name gpt2 --use_tril_attention_mask
 python lit_export.py --version 0
 python generate.py --model_name_or_path exports/version_0 --tokenizer_name_or_path gpt2
 ```
+> :memo: **Note:** Training with a "--use_tril_attention_mask" is recommended. However, huggingface model implementions might not support 2D attention mask. You may write a custom model to support 2D attention mask, just like what I did in [custom_models/gpt2](https://github.com/Yiqing-Zhou/gpt-pretrain/tree/main/custom_models/gpt2).
+
+## Train on multiple GPUs
+
+```
+python lit_train.py --model_name gpt2 --use_tril_attention_mask --strategy fsdp # default and recommended
+```
+
+```
+python lit_train.py --model_name gpt2 --use_tril_attention_mask --strategy deepspeed
+```
+
+```
+python lit_train.py --model_name gpt2 --use_tril_attention_mask --strategy ddp
+```
+
+## Reduce CUDA memory cost
+
+- half precision
+    ```
+    python lit_train.py --model_name gpt2 --use_tril_attention_mask --bf16
+    ```
+    ```
+    python lit_train.py --model_name gpt2 --use_tril_attention_mask --fp16
+    ```
+- smaller batch size & accumulate grad batches
+    ```
+    python lit_train.py --model_name gpt2 --use_tril_attention_mask --bf16 \
+        --train_batch_size 2 --val_batch_size 4 --accumulate_grad_batches 128
+    ```
+- cpu_offload
+    ```
+    python lit_train.py --model_name gpt2 --use_tril_attention_mask --bf16 \
+        --strategy fsdp_cpu_offload
+    ```
+    ```
+    python lit_train.py --model_name gpt2 --use_tril_attention_mask --bf16 \
+        --strategy deepspeed_stage_3_offload
+    ```
+    

custom_models/__init__.py

@@ -0,0 +1,60 @@
+import importlib
+from collections import OrderedDict
+
+from transformers.models.auto import auto_factory, configuration_auto
+
+CONFIG_MAPPING_NAMES = OrderedDict([])
+
+
+def register_custom_configs():
+    for model_type, map_name in CONFIG_MAPPING_NAMES.items():
+        module_name = configuration_auto.model_type_to_module_name(model_type)
+        module = importlib.import_module(f".{module_name}", "custom_models")
+        mapping = getattr(module, map_name)
+        configuration_auto.AutoConfig.register(model_type, mapping)
+
+
+class _LazyAutoMapping(auto_factory._LazyAutoMapping):
+    def _load_attr_from_module(self, model_type, attr):
+        module_name = auto_factory.model_type_to_module_name(model_type)
+        if module_name not in self._modules:
+            self._modules[module_name] = importlib.import_module(f".{module_name}", "custom_models")
+        return auto_factory.getattribute_from_module(self._modules[module_name], attr)
+
+
+MODEL_MAPPING_NAMES = OrderedDict(
+    [
+        ("gpt2", "GPT2Model"),
+    ]
+)
+
+
+MODEL_FOR_CAUSAL_LM_MAPPING_NAMES = OrderedDict(
+    [
+        ("gpt2", "GPT2LMHeadModel"),
+    ]
+)
+
+
+MODEL_MAPPING = _LazyAutoMapping(
+    {**configuration_auto.CONFIG_MAPPING_NAMES, **CONFIG_MAPPING_NAMES}, MODEL_MAPPING_NAMES
+)
+
+
+MODEL_FOR_CAUSAL_LM_MAPPING = _LazyAutoMapping(
+    {**configuration_auto.CONFIG_MAPPING_NAMES, **CONFIG_MAPPING_NAMES}, MODEL_FOR_CAUSAL_LM_MAPPING_NAMES
+)
+
+
+class AutoModel(auto_factory._BaseAutoModelClass):
+    _model_mapping = MODEL_MAPPING
+
+
+AutoModel = auto_factory.auto_class_update(AutoModel)
+
+
+class AutoModelForCausalLM(auto_factory._BaseAutoModelClass):
+    _model_mapping = MODEL_FOR_CAUSAL_LM_MAPPING
+
+
+AutoModelForCausalLM = auto_factory.auto_class_update(AutoModelForCausalLM)

custom_models/gpt2/__init__.py

@@ -0,0 +1 @@
+from .modeling_gpt2 import GPT2LMHeadModel, GPT2Model

custom_models/gpt2/modeling_gpt2.py

@@ -0,0 +1,355 @@
+"""Override transformers GPT2 to support tril attention mask"""
+
+from typing import Any, Dict, Optional, Tuple, Union
+
+import torch
+import torch.utils.checkpoint
+import transformers
+from torch import nn
+from transformers.modeling_outputs import BaseModelOutputWithPastAndCrossAttentions
+from transformers.models.gpt2.modeling_gpt2 import (
+    _CHECKPOINT_FOR_DOC,
+    _CONFIG_FOR_DOC,
+    GPT2_INPUTS_DOCSTRING,
+    logger,
+)
+from transformers.utils import (
+    ModelOutput,
+    add_code_sample_docstrings,
+    add_start_docstrings_to_model_forward,
+)
+
+
+class GPT2Model(transformers.models.gpt2.GPT2Model):
+    @add_start_docstrings_to_model_forward(GPT2_INPUTS_DOCSTRING)
+    @add_code_sample_docstrings(
+        checkpoint=_CHECKPOINT_FOR_DOC,
+        output_type=BaseModelOutputWithPastAndCrossAttentions,
+        config_class=_CONFIG_FOR_DOC,
+    )
+    def forward(
+        self,
+        input_ids: Optional[torch.LongTensor] = None,
+        past_key_values: Optional[Tuple[Tuple[torch.Tensor]]] = None,
+        attention_mask: Optional[torch.FloatTensor] = None,
+        token_type_ids: Optional[torch.LongTensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        head_mask: Optional[torch.FloatTensor] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        encoder_hidden_states: Optional[torch.Tensor] = None,
+        encoder_attention_mask: Optional[torch.FloatTensor] = None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+    ) -> Union[Tuple, BaseModelOutputWithPastAndCrossAttentions]:
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        output_hidden_states = (
+            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        )
+        use_cache = use_cache if use_cache is not None else self.config.use_cache
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        if input_ids is not None and inputs_embeds is not None:
+            raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
+        elif input_ids is not None:
+            input_shape = input_ids.size()
+            input_ids = input_ids.view(-1, input_shape[-1])
+            batch_size = input_ids.shape[0]
+        elif inputs_embeds is not None:
+            input_shape = inputs_embeds.size()[:-1]
+            batch_size = inputs_embeds.shape[0]
+        else:
+            raise ValueError("You have to specify either input_ids or inputs_embeds")
+
+        device = input_ids.device if input_ids is not None else inputs_embeds.device
+
+        if token_type_ids is not None:
+            token_type_ids = token_type_ids.view(-1, input_shape[-1])
+        if position_ids is not None:
+            position_ids = position_ids.view(-1, input_shape[-1])
+
+        if past_key_values is None:
+            past_length = 0
+            past_key_values = tuple([None] * len(self.h))
+        else:
+            past_length = past_key_values[0][0].size(-2)
+        if position_ids is None:
+            position_ids = torch.arange(
+                past_length,
+                input_shape[-1] + past_length,
+                dtype=torch.long,
+                device=device,
+            )
+            position_ids = position_ids.unsqueeze(0).view(-1, input_shape[-1])
+
+        # GPT2Attention mask.
+        if attention_mask is not None:
+            if batch_size <= 0:
+                raise ValueError("batch_size has to be defined and > 0")
+            if attention_mask.dim() == 2:
+                # We create a 3D attention mask from a 2D tensor mask.
+                # Sizes are [batch_size, 1, 1, to_seq_length]
+                # So we can broadcast to [batch_size, num_heads, from_seq_length, to_seq_length]
+                # this attention mask is more simple than the triangular masking of causal attention
+                # used in OpenAI GPT, we just need to prepare the broadcast dimension here.
+                attention_mask = attention_mask[:, None, None, :]
+            elif attention_mask.dim() == 3:
+                attention_mask = attention_mask[:, None, ...]
+            else:
+                raise ValueError(f"attention_mask.dim() is {attention_mask.dim()}, should be 2 or 3")
+
+            # Since attention_mask is 1.0 for positions we want to attend and 0.0 for
+            # masked positions, this operation will create a tensor which is 0.0 for
+            # positions we want to attend and the dtype's smallest value for masked positions.
+            # Since we are adding it to the raw scores before the softmax, this is
+            # effectively the same as removing these entirely.
+            attention_mask = attention_mask.to(dtype=self.dtype)  # fp16 compatibility
+            attention_mask = (1.0 - attention_mask) * torch.finfo(self.dtype).min
+
+        # If a 2D or 3D attention mask is provided for the cross-attention
+        # we need to make broadcastable to [batch_size, num_heads, seq_length, seq_length]
+        if self.config.add_cross_attention and encoder_hidden_states is not None:
+            (
+                encoder_batch_size,
+                encoder_sequence_length,
+                _,
+            ) = encoder_hidden_states.size()
+            encoder_hidden_shape = (encoder_batch_size, encoder_sequence_length)
+            if encoder_attention_mask is None:
+                encoder_attention_mask = torch.ones(encoder_hidden_shape, device=device)
+            encoder_attention_mask = self.invert_attention_mask(encoder_attention_mask)
+        else:
+            encoder_attention_mask = None
+
+        # Prepare head mask if needed
+        # 1.0 in head_mask indicate we keep the head
+        # attention_probs has shape bsz x n_heads x N x N
+        # head_mask has shape n_layer x batch x n_heads x N x N
+        head_mask = self.get_head_mask(head_mask, self.config.n_layer)
+
+        if inputs_embeds is None:
+            inputs_embeds = self.wte(input_ids)
+        position_embeds = self.wpe(position_ids)
+        hidden_states = inputs_embeds + position_embeds
+
+        if token_type_ids is not None:
+            token_type_embeds = self.wte(token_type_ids)
+            hidden_states = hidden_states + token_type_embeds
+
+        hidden_states = self.drop(hidden_states)
+
+        output_shape = input_shape + (hidden_states.size(-1),)
+
+        if self.gradient_checkpointing and self.training:
+            if use_cache:
+                logger.warning_once(
+                    "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
+                )
+                use_cache = False
+
+        presents = () if use_cache else None
+        all_self_attentions = () if output_attentions else None
+        all_cross_attentions = () if output_attentions and self.config.add_cross_attention else None
+        all_hidden_states = () if output_hidden_states else None
+        for i, (block, layer_past) in enumerate(zip(self.h, past_key_values)):
+            # Model parallel
+            if self.model_parallel:
+                torch.cuda.set_device(hidden_states.device)
+                # Ensure layer_past is on same device as hidden_states (might not be correct)
+                if layer_past is not None:
+                    layer_past = tuple(past_state.to(hidden_states.device) for past_state in layer_past)
+                # Ensure that attention_mask is always on the same device as hidden_states
+                if attention_mask is not None:
+                    attention_mask = attention_mask.to(hidden_states.device)
+                if isinstance(head_mask, torch.Tensor):
+                    head_mask = head_mask.to(hidden_states.device)
+            if output_hidden_states:
+                all_hidden_states = all_hidden_states + (hidden_states,)
+
+            if self.gradient_checkpointing and self.training:
+
+                def create_custom_forward(module):
+                    def custom_forward(*inputs):
+                        # None for past_key_value
+                        return module(*inputs, use_cache, output_attentions)
+
+                    return custom_forward
+
+                outputs = torch.utils.checkpoint.checkpoint(
+                    create_custom_forward(block),
+                    hidden_states,
+                    None,
+                    attention_mask,
+                    head_mask[i],
+                    encoder_hidden_states,
+                    encoder_attention_mask,
+                )
+            else:
+                outputs = block(
+                    hidden_states,
+                    layer_past=layer_past,
+                    attention_mask=attention_mask,
+                    head_mask=head_mask[i],
+                    encoder_hidden_states=encoder_hidden_states,
+                    encoder_attention_mask=encoder_attention_mask,
+                    use_cache=use_cache,
+                    output_attentions=output_attentions,
+                )
+
+            hidden_states = outputs[0]
+            if use_cache is True:
+                presents = presents + (outputs[1],)
+
+            if output_attentions:
+                all_self_attentions = all_self_attentions + (outputs[2 if use_cache else 1],)
+                if self.config.add_cross_attention:
+                    all_cross_attentions = all_cross_attentions + (outputs[3 if use_cache else 2],)
+
+            # Model Parallel: If it's the last layer for that device, put things on the next device
+            if self.model_parallel:
+                for k, v in self.device_map.items():
+                    if i == v[-1] and "cuda:" + str(k) != self.last_device:
+                        hidden_states = hidden_states.to("cuda:" + str(k + 1))
+
+        hidden_states = self.ln_f(hidden_states)
+
+        hidden_states = hidden_states.view(output_shape)
+        # Add last hidden state
+        if output_hidden_states:
+            all_hidden_states = all_hidden_states + (hidden_states,)
+
+        if not return_dict:
+            return tuple(
+                v
+                for v in [
+                    hidden_states,
+                    presents,
+                    all_hidden_states,
+                    all_self_attentions,
+                    all_cross_attentions,
+                ]
+                if v is not None
+            )
+
+        return BaseModelOutputWithPastAndCrossAttentions(
+            last_hidden_state=hidden_states,
+            past_key_values=presents,
+            hidden_states=all_hidden_states,
+            attentions=all_self_attentions,
+            cross_attentions=all_cross_attentions,
+        )
+
+
+class GPT2LMHeadModel(transformers.models.gpt2.GPT2LMHeadModel):
+    def __init__(self, config):
+        super().__init__(config)
+        self.transformer = GPT2Model(config)
+        self.lm_head = nn.Linear(config.n_embd, config.vocab_size, bias=False)
+
+        # Model parallel
+        self.model_parallel = False
+        self.device_map = None
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def prepare_inputs_for_generation(self, input_ids, past_key_values=None, inputs_embeds=None, **kwargs):
+        token_type_ids = kwargs.get("token_type_ids", None)
+        # only last token for inputs_ids if past is defined in kwargs
+        if past_key_values:
+            input_ids = input_ids[:, -1].unsqueeze(-1)
+            if token_type_ids is not None:
+                token_type_ids = token_type_ids[:, -1].unsqueeze(-1)
+
+        attention_mask = kwargs.get("attention_mask", None)
+        position_ids = kwargs.get("position_ids", None)
+
+        if attention_mask is not None and position_ids is None:
+            # create position_ids on the fly for batch generation
+            position_ids = attention_mask.long().cumsum(-1) - 1
+            position_ids.masked_fill_(attention_mask == 0, 1)
+            if past_key_values:
+                position_ids = position_ids[:, -1].unsqueeze(-1)
+
+        # if `inputs_embeds` are passed, we only want to use them in the 1st generation step
+        if inputs_embeds is not None and past_key_values is None:
+            model_inputs = {"inputs_embeds": inputs_embeds}
+        else:
+            model_inputs = {"input_ids": input_ids}
+
+        model_inputs.update(
+            {
+                "past_key_values": past_key_values,
+                "use_cache": kwargs.get("use_cache"),
+                "position_ids": position_ids,
+                "attention_mask": attention_mask,
+                "token_type_ids": token_type_ids,
+            }
+        )
+        return model_inputs
+
+    def _update_model_kwargs_for_generation(
+        self,
+        outputs: ModelOutput,
+        model_kwargs: Dict[str, Any],
+        is_encoder_decoder: bool = False,
+        standardize_cache_format: bool = False,
+    ) -> Dict[str, Any]:
+        # update past_key_values
+        model_kwargs["past_key_values"] = self._extract_past_from_model_output(
+            outputs, standardize_cache_format=standardize_cache_format
+        )
+
+        # update token_type_ids with last value
+        if "token_type_ids" in model_kwargs:
+            token_type_ids = model_kwargs["token_type_ids"]
+            model_kwargs["token_type_ids"] = torch.cat([token_type_ids, token_type_ids[:, -1].unsqueeze(-1)], dim=-1)
+
+        # update position_ids
+        if "position_ids" in model_kwargs:
+            position_ids = model_kwargs["position_ids"]
+            if model_kwargs["past_key_values"] is not None:
+                model_kwargs["position_ids"] = (position_ids[:, -1] + 1).unsqueeze(-1)
+            else:
+                model_kwargs["position_ids"] = torch.cat(
+                    [position_ids, (position_ids[:, -1] + 1).unsqueeze(-1)], dim=-1
+                )
+
+        if not is_encoder_decoder:
+            # update attention mask
+            if "attention_mask" in model_kwargs:
+                attention_mask = model_kwargs["attention_mask"]
+                if attention_mask.dim() == 2:
+                    model_kwargs["attention_mask"] = torch.cat(
+                        [
+                            attention_mask,
+                            attention_mask.new_ones((attention_mask.shape[0], 1)),
+                        ],
+                        dim=-1,
+                    )
+                elif attention_mask.dim() == 3:
+                    attention_mask = attention_mask[:, -1, :]
+                    attention_mask = torch.cat(
+                        [
+                            attention_mask,
+                            attention_mask.new_ones((attention_mask.shape[0], 1)),
+                        ],
+                        dim=-1,
+                    )
+                    model_kwargs["attention_mask"] = attention_mask
+                else:
+                    raise ValueError(f"attention_mask.dim() is {attention_mask.dim()}, should be 2 or 3")
+        else:
+            # update decoder attention mask
+            if "decoder_attention_mask" in model_kwargs:
+                decoder_attention_mask = model_kwargs["decoder_attention_mask"]
+                model_kwargs["decoder_attention_mask"] = torch.cat(
+                    [
+                        decoder_attention_mask,
+                        decoder_attention_mask.new_ones((decoder_attention_mask.shape[0], 1)),
+                    ],
+                    dim=-1,
+                )
+
+        return model_kwargs

generate.py

@@ -13,15 +13,11 @@ def eval_prompts(
     prompts: List[str],
     use_tril_attention_mask: bool = False,
 ) -> List[str]:
-    inputs = tokenizer(
+    inputs = tokenizer(prompts, padding=True, return_tensors='pt', return_attention_mask=True)
-        prompts, padding=True, return_tensors='pt', return_attention_mask=True
-    )
     inputs['position_ids'] = inputs.attention_mask.cumsum(-1) - 1
     inputs['position_ids'].masked_fill_(inputs.attention_mask == 0, 1)
     if use_tril_attention_mask:
-        inputs['attention_mask'] = (
+        inputs['attention_mask'] = (inputs.attention_mask.unsqueeze(1) * inputs.attention_mask.unsqueeze(2)).tril()
-            inputs.attention_mask.unsqueeze(1) * inputs.attention_mask.unsqueeze(2)
-        ).tril()
     inputs = inputs.to(model.device)
     with torch.inference_mode():
         output_ids = model.generate(
@@ -32,9 +28,7 @@ def eval_prompts(
             eos_token_id=tokenizer.eos_token_id,
             early_stopping=True,
         )
-    completes = tokenizer.batch_decode(
+    completes = tokenizer.batch_decode(output_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)
-        output_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False
-    )
     return completes
 
 
@@ -81,9 +75,7 @@ if __name__ == '__main__':
         "这是一个最好的时代，这是一个最坏的时代。",
         "这是一个最好的时代，这是一个最坏的",
     ]
-    completes = eval_prompts(
+    completes = eval_prompts(model, tokenizer, prompts, use_tril_attention_mask=args.use_tril_attention_mask)
-        model, tokenizer, prompts, use_tril_attention_mask=args.use_tril_attention_mask
-    )
 
     for prompt, complete in zip(prompts, completes):
         print("[p]", prompt)

lit_export.py

@@ -26,8 +26,6 @@ if __name__ == '__main__':
 
     checkpoint_file_path = next(lightning_logs_dir_path.glob("checkpoints/*.ckpt"))
 
-    lit_module = LitModule.load_from_checkpoint(
+    lit_module = LitModule.load_from_checkpoint(checkpoint_file_path, map_location='cpu')
-        checkpoint_file_path, map_location='cpu'
-    )
     model: PreTrainedModel = lit_module.__core_module__
     model.save_pretrained(exports_dir_path)

lit_module.py

@@ -27,9 +27,7 @@ class LitModule(pl.LightningModule):
         )
 
     @cache
-    def get_batch_tril_matrix(
+    def get_batch_tril_matrix(self, block_size: int, batch_size: Optional[int] = None) -> torch.Tensor:
-        self, block_size: int, batch_size: Optional[int] = None
-    ) -> torch.Tensor:
         matrix = torch.ones(block_size, block_size).tril()
         if batch_size is not None:
             matrix = matrix.repeat(batch_size, 1, 1)
@@ -42,9 +40,7 @@ class LitModule(pl.LightningModule):
     def training_step(self, batch: Dict[str, torch.Tensor], batch_idx):
         batch_size, block_size = batch['input_ids'].shape
         if self.use_tril_attention_mask:
-            batch['attention_mask'] = self.get_batch_tril_matrix(
+            batch['attention_mask'] = self.get_batch_tril_matrix(block_size, batch_size=batch_size).to(self.device)
-                block_size, batch_size=batch_size
-            ).to(self.device)
         outputs = self.llm(**batch, return_dict=True)
         loss = outputs.loss
 
@@ -80,9 +76,7 @@ class LitModule(pl.LightningModule):
                         self.trainer.model.parameters(), lr=self.learning_rate
                     )
                     return optimizer
-        optimizer = torch.optim.AdamW(
+        optimizer = torch.optim.AdamW(self.trainer.model.parameters(), lr=self.learning_rate)
-            self.trainer.model.parameters(), lr=self.learning_rate
-        )
         return optimizer
 
     def configure_callbacks(self):

lit_train.py

@@ -25,16 +25,12 @@ def split_raw_dataset(
     if 'validation' in raw_dataset:
         train_dataset, val_dataset = raw_dataset['train'], raw_dataset['validation']
     else:
-        raw_dataset = raw_dataset['train'].train_test_split(
+        raw_dataset = raw_dataset['train'].train_test_split(test_size=0.05, seed=args.seed)
-            test_size=0.05, seed=args.seed
-        )
         train_dataset, val_dataset = raw_dataset['train'], raw_dataset['test']
     return train_dataset, val_dataset
 
 
-def process_dataset(
+def process_dataset(dataset: datasets.Dataset, tokenizer: PreTrainedTokenizer) -> datasets.Dataset:
-    dataset: datasets.Dataset, tokenizer: PreTrainedTokenizer
-) -> datasets.Dataset:
     def group_texts(examples: Dict[str, list], block_size: int = 512) -> BatchEncoding:
         concatenated_examples = {k: list(chain(*examples[k])) for k in examples.keys()}
         total_length = len(concatenated_examples[list(examples.keys())[0]])
@@ -110,13 +106,13 @@ def parse_args():
         "--train_batch_size",
         type=int,
         help="Batch size of training",
-        default=8,
+        default=2,
     )
     parser.add_argument(
         "--val_batch_size",
         type=int,
         help="Batch size of validating",
-        default=16,
+        default=2,
     )
     parser.add_argument(
         "--accumulate_grad_batches",
@@ -128,7 +124,7 @@ def parse_args():
         "--num_proc",
         type=str,
         help="Number of data processes",
-        default=16,
+        default=1,
     )
     parser.add_argument(
         "--max_epochs",
@@ -140,7 +136,7 @@ def parse_args():
         "--strategy",
         type=str,
         help="Name of pytorch lightning distribution strategy",
-        default='fsdp',
+        default='ddp',
     )
     parser.add_argument(
         "--resume_from_ckpt_path",
@@ -167,9 +163,7 @@ if __name__ == '__main__':
     set_seed(args.seed)
 
     # lightning module
-    lit_module = LitModule(
+    lit_module = LitModule(args.model_name, args.learning_rate, args.use_tril_attention_mask)
-        args.model_name, args.learning_rate, args.use_tril_attention_mask
-    )
 
     # datasets
     tokenizer = load_tokenizer(args.tokenizer_name_or_path)
@@ -203,8 +197,11 @@ if __name__ == '__main__':
         persistent_workers=True,
     )
 
+    ne = next(train_dataloader._get_iterator())
+    print((ne["input_ids"]-ne["labels"]).numpy().tolist())
+
     # trainer
-    apply_all_patches()
+    # apply_all_patches()
     torch.set_float32_matmul_precision('medium')
     if args.bf16:
         precision = 'bf16-mixed'

utils.py

@@ -10,9 +10,19 @@ from transformers import (
     PreTrainedTokenizer,
 )
 
+import custom_models
 
-def init_model(model_name: Union[str, os.PathLike]) -> PreTrainedModel:
+custom_models.register_custom_configs()
-    config = AutoConfig.from_pretrained(model_name, trust_remote_code=True)
+
+
+def init_model(model_name: str) -> PreTrainedModel:
+    config = AutoConfig.for_model(model_type=model_name)
+
+    if model_name in custom_models.MODEL_FOR_CAUSAL_LM_MAPPING_NAMES:
+        model = custom_models.AutoModelForCausalLM.from_config(config)
+    elif model_name in custom_models.MODEL_MAPPING_NAMES:
+        model = custom_models.AutoModel.from_config(config)
+    else:
         try:
             model = AutoModelForCausalLM.from_config(config, trust_remote_code=True)
         except ValueError:
@@ -21,21 +31,22 @@ def init_model(model_name: Union[str, os.PathLike]) -> PreTrainedModel:
 
 
 def load_model(model_name_or_path: Union[str, os.PathLike]) -> PreTrainedModel:
+    config = AutoConfig.from_pretrained(model_name_or_path)
+
+    if config.model_type in custom_models.MODEL_FOR_CAUSAL_LM_MAPPING_NAMES:
+        model = custom_models.AutoModelForCausalLM.from_pretrained(model_name_or_path, config=config)
+    elif config.model_type in custom_models.MODEL_MAPPING_NAMES:
+        model = custom_models.AutoModel.from_pretrained(model_name_or_path, config=config)
+    else:
         try:
-        model = AutoModelForCausalLM.from_pretrained(
+            model = AutoModelForCausalLM.from_pretrained(model_name_or_path, config=config, trust_remote_code=True)
-            model_name_or_path, trust_remote_code=True
-        )
         except ValueError:
-        model = AutoModel.from_pretrained(model_name_or_path, trust_remote_code=True)
+            model = AutoModel.from_pretrained(model_name_or_path, config=config, trust_remote_code=True)
     return model
 
 
-def load_tokenizer(
+def load_tokenizer(tokenizer_name_or_path: Union[str, os.PathLike]) -> PreTrainedTokenizer:
-    tokenizer_name_or_path: Union[str, os.PathLike]
+    tokenizer = AutoTokenizer.from_pretrained(tokenizer_name_or_path, padding_side='left', trust_remote_code=True)
-) -> PreTrainedTokenizer:
-    tokenizer = AutoTokenizer.from_pretrained(
-        tokenizer_name_or_path, padding_side='left', trust_remote_code=True
-    )
     if tokenizer.pad_token_id is None:
         tokenizer.pad_token = tokenizer.eos_token
     return tokenizer