fix

examples/pytorch/llm/scripts/c4ai_command_r_plus/lora_mp/infer.sh

@@ -3,6 +3,7 @@
 CUDA_VISIBLE_DEVICES=0,1,2,3 \
 swift infer \
     --ckpt_dir "output/c4ai-command-r-plus/vx-xxx/checkpoint-xx" \
+    --load_dataset_config true \
     --load_args_from_ckpt_dir true \
     --temperature 0.3 \
     --top_p 0.7 \

examples/pytorch/llm/scripts/c4ai_command_r_plus/lora_mp/sft.sh

@@ -29,4 +29,4 @@ swift sft \
     --save_steps 100 \
     --save_total_limit 2 \
     --logging_steps 10 \
-    --use_flash_attn true \
+    --use_flash_attn true \

swift/llm/utils/model.py

@@ -455,55 +455,6 @@ def _register_model(
     TemplateType.mengzi,
     support_vllm=True,
     support_flash_attn=True)
-# @register_model(
-#     ModelType.c4ai_command_r_v01,
-#     'AI-ModelScope/c4ai-command-r-v01',
-#     LoRATM.llama2,
-#     TemplateType.c4ai,
-#     requires=['transformers>=4.39.1'],
-#     support_vllm=True,
-#     support_flash_attn=True)
-# @register_model(
-#     ModelType.c4ai_command_r_plus,
-#     'AI-ModelScope/c4ai-command-r-plus',
-#     LoRATM.llama2,
-#     TemplateType.c4ai,
-#     requires=['transformers>4.39'],
-#     support_vllm=True,
-#     support_flash_attn=True)
-def get_model_tokenizer_from_repo(model_dir: str,
-                                  torch_dtype: Optional[Dtype],
-                                  model_kwargs: Dict[str, Any],
-                                  load_model: bool = True,
-                                  model_config=None,
-                                  tokenizer=None,
-                                  automodel_class=AutoModelForCausalLM,
-                                  **kwargs):
-    """load from an independent repository"""
-    if model_config is None:
-        model_config = AutoConfig.from_pretrained(
-            model_dir, trust_remote_code=True)
-    if torch_dtype is not None:
-        model_config.torch_dtype = torch_dtype
-    if tokenizer is None:
-        tokenizer = AutoTokenizer.from_pretrained(
-            model_dir, trust_remote_code=True)
-    eos_token = kwargs.get('eos_token')
-    if eos_token is not None:
-        tokenizer.eos_token = eos_token
-    model = None
-    context = kwargs.get('context', nullcontext())
-    if load_model:
-        with context:
-            model = automodel_class.from_pretrained(
-                model_dir,
-                config=model_config,
-                torch_dtype=torch_dtype,
-                trust_remote_code=True,
-                **model_kwargs)
-    return model, tokenizer
-
-
 @register_model(
     ModelType.c4ai_command_r_v01,
     'AI-ModelScope/c4ai-command-r-v01',
@@ -520,22 +471,23 @@ def get_model_tokenizer_from_repo(model_dir: str,
     requires=['transformers>4.39'],
     support_vllm=False,
     support_flash_attn=True)
-def get_model_tokenizer_c4ai(model_dir: str,
-                             torch_dtype: Optional[Dtype],
-                             model_kwargs: Dict[str, Any],
-                             load_model: bool = True,
-                             model_config=None,
-                             tokenizer=None,
-                             automodel_class=AutoModelForCausalLM,
-                             **kwargs):
+def get_model_tokenizer_from_repo(model_dir: str,
+                                  torch_dtype: Optional[Dtype],
+                                  model_kwargs: Dict[str, Any],
+                                  load_model: bool = True,
+                                  model_config=None,
+                                  tokenizer=None,
+                                  automodel_class=AutoModelForCausalLM,
+                                  **kwargs):
+    """load from an independent repository"""
     if model_config is None:
         model_config = AutoConfig.from_pretrained(
             model_dir, trust_remote_code=True)
     if torch_dtype is not None:
         model_config.torch_dtype = torch_dtype
     if tokenizer is None:
         tokenizer = AutoTokenizer.from_pretrained(
-            model_dir, trust_remote_code=True, use_fast=False)
+            model_dir, trust_remote_code=True)
     eos_token = kwargs.get('eos_token')
     if eos_token is not None:
         tokenizer.eos_token = eos_token
@@ -547,6 +499,7 @@ def get_model_tokenizer_c4ai(model_dir: str,
                 model_dir,
                 config=model_config,
                 torch_dtype=torch_dtype,
+                trust_remote_code=True,
                 **model_kwargs)
     return model, tokenizer