fix issue in scripts (#3456)

Author: blzheng

examples/cpu/llm/inference/distributed/run_generation_with_deepspeed.py

@@ -328,8 +328,8 @@ def get_checkpoint_files(model_name_or_path):
 model_type = next((x for x in MODEL_CLASSES.keys() if x in model_name.lower()), "auto")
 if model_type == "llama" and args.vision_text_model:
     model_type = "mllama"
-if model_type == "maira-2":
-    model_type = "maira2"
+if model_type in ["maira-2", "deepseek-v2", "deepseek-v3"]:
+    model_type = model_type.replace("-", "")
 model_class = MODEL_CLASSES[model_type]
 tokenizer = model_class[1].from_pretrained(model_name, trust_remote_code=True)
 

examples/cpu/llm/inference/single_instance/run_generation.py

@@ -140,8 +140,8 @@
 )
 if model_type == "llama" and args.vision_text_model:
     model_type = "mllama"
-if model_type == "maira-2":
-    model_type = "maira2"
+if model_type in ["maira-2", "deepseek-v2", "deepseek-v3"]:
+    model_type = model_type.replace("-", "")
 model_class = MODEL_CLASSES[model_type]
 if args.config_file is None:
     if model_type == "chatglm":

examples/cpu/llm/inference/utils/create_shard_model.py

@@ -53,8 +53,8 @@
     )
     if model_type == "llama" and args.vision_text_model:
         model_type = "mllama"
-    if model_type == "maira-2":
-        model_type = "maira2"
+    if model_type in ["maira-2", "deepseek-v2", "deepseek-v3"]:
+        model_type = model_type.replace("-", "")
     model_class = MODEL_CLASSES[model_type]
     load_dtype = torch.float32
     if args.dtype == "float16":
@@ -83,7 +83,7 @@
     tokenizer.save_pretrained(save_directory=args.save_path)
     if model_type == "llava":
         image_processor.save_pretrained(save_directory=args.save_path)
-    if model_type == "maira2":
+    if model_type in ["maira2", "deepseekv2", "deepseekv3"]:
         import inspect
         import shutil
 

examples/cpu/llm/inference/utils/supported_models.py

@@ -37,6 +37,8 @@
     "jamba": (AutoModelForCausalLM, AutoTokenizer),
     "deepseek-v2": (AutoModelForCausalLM, AutoTokenizer),
     "deepseek-v3": (AutoModelForCausalLM, AutoTokenizer),
+    "deepseekv2": (AutoModelForCausalLM, AutoTokenizer),
+    "deepseekv3": (AutoModelForCausalLM, AutoTokenizer),
     "auto": (AutoModelForCausalLM, AutoTokenizer),
 }