remove flashmask checker (#2631)

Author: WYB27

examples/run_finetune.py

@@ -37,26 +37,8 @@
     AutoModelForCausalLM,
     AutoModelForCausalLMPipe,
     AutoTokenizer,
-    DeepseekV2ForCausalLM,
-    DeepseekV2ForCausalLMPipe,
-    DeepseekV3ForCausalLM,
-    DeepseekV3ForCausalLMPipe,
-    Ernie4_5_MoeForCausalLM,
-    Ernie4_5_MoeForCausalLMPipe,
-    Ernie4_5ForCausalLM,
-    Ernie4_5ForCausalLMPipe,
     Llama3Tokenizer,
-    LlamaForCausalLM,
-    LlamaForCausalLMPipe,
     LlamaTokenizer,
-    Qwen2ForCausalLM,
-    Qwen2ForCausalLMPipe,
-    Qwen2MoeForCausalLM,
-    Qwen2MoeForCausalLMPipe,
-    Qwen3ForCausalLM,
-    Qwen3ForCausalLMPipe,
-    Qwen3MoeForCausalLM,
-    Qwen3MoeForCausalLMPipe,
 )
 from paddleformers.transformers.configuration_utils import LlmMetaConfig
 from paddleformers.trl import DataConfig, ModelConfig, SFTConfig, SFTTrainer
@@ -66,27 +48,6 @@
 # Fine-tune Environment Variables to support sharding stage1 overlap optimization.
 os.environ["USE_CASUAL_MASK"] = "False"
 
-flash_mask_support_list = [
-    DeepseekV2ForCausalLM,
-    DeepseekV2ForCausalLMPipe,
-    DeepseekV3ForCausalLM,
-    DeepseekV3ForCausalLMPipe,
-    Ernie4_5ForCausalLM,
-    Ernie4_5ForCausalLMPipe,
-    Ernie4_5_MoeForCausalLM,
-    Ernie4_5_MoeForCausalLMPipe,
-    LlamaForCausalLM,
-    LlamaForCausalLMPipe,
-    Qwen2ForCausalLM,
-    Qwen2ForCausalLMPipe,
-    Qwen2MoeForCausalLM,
-    Qwen2MoeForCausalLMPipe,
-    Qwen3ForCausalLM,
-    Qwen3ForCausalLMPipe,
-    Qwen3MoeForCausalLM,
-    Qwen3MoeForCausalLMPipe,
-]
-
 
 def main():
     parser = PdArgumentParser((ModelConfig, DataConfig, SFTConfig))
@@ -192,9 +153,6 @@ def main():
     else:
         model = model_class.from_config(model_config, dtype=dtype)
 
-    if model_args.attn_impl == "flashmask" and not any(isinstance(model, cls) for cls in flash_mask_support_list):
-        raise NotImplementedError(f"{model.__class__} not support flash mask.")
-
     if training_args.do_train and model_args.neftune:
         # Inspired by https://github.com/neelsjain/NEFTune
         if hasattr(model, "get_input_embeddings"):