[Device]Support npu (#6159)

* support npu

* support pretrain

support pretrain

fix

* support lora

fix

fix

* support chatglm

fix

fxi

fix

[pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

fix

fix

[pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

fix

[pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

fix

fix

fix

* Update train.py

* Update train.py

* fix

* fix

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* fix

* fix

* fix

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

---------

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>

applications/Colossal-LLaMA/colossal_llama/dataset/conversation.py

@@ -100,7 +100,7 @@ LLaMA3_Conv = Conversation(
     messages=[],
     offset=0,
     sep_style=SeparatorStyle.ADD_BOS_EOS_TOKEN,
-    seps=["<|begin_of_text|>", "<|end_of_text|>"],
+    seps=["<|begin_of_text|>", "<|eot_id|>"],
 )
 
 default_conversation = LLaMA3_Conv

applications/Colossal-LLaMA/colossal_llama/dataset/spliced_and_tokenized_dataset.py

@@ -88,7 +88,7 @@ def supervised_tokenize_sft(
 
     assert (
         tokenizer.bos_token == conversation_template.seps[0] and tokenizer.eos_token == conversation_template.seps[1]
-    ), "`bos_token` and `eos_token` should be the same with `conversation_template.seps`."
+    ), f"`bos_token`{tokenizer.bos_token} and `eos_token`{tokenizer.eos_token} should be the same with `conversation_template.seps`{conversation_template.seps}."
 
     if ignore_index is None:
         ignore_index = IGNORE_INDEX

applications/Colossal-LLaMA/colossal_llama/utils/ckpt_io.py

@@ -43,6 +43,7 @@ def save_checkpoint(
     step: int,
     batch_size: int,
     coordinator: DistCoordinator,
+    use_lora: bool = False,
 ) -> None:
     """
     Save model checkpoint, optimizer, LR scheduler and intermedidate running states.
@@ -51,7 +52,10 @@ def save_checkpoint(
     save_dir = os.path.join(save_dir, f"epoch-{epoch}_step-{step}")
     os.makedirs(os.path.join(save_dir, "modeling"), exist_ok=True)
 
-    booster.save_model(model, os.path.join(save_dir, "modeling"), shard=True)
+    if use_lora:
+        booster.save_lora_as_pretrained(model, os.path.join(save_dir, "modeling"))
+    else:
+        booster.save_model(model, os.path.join(save_dir, "modeling"), shard=True)
 
     booster.save_optimizer(optimizer, os.path.join(save_dir, "optimizer"), shard=True)
     booster.save_lr_scheduler(lr_scheduler, os.path.join(save_dir, "lr_scheduler"))