model: Minicpmo (#3023)

benchmark/mmmu/bench_hf.py

@@ -1,5 +1,6 @@
 import argparse
 
+import PIL.Image
 import torch
 from data_utils import save_json
 from eval_utils import (
@@ -10,22 +11,38 @@ from eval_utils import (
     process_result,
 )
 from tqdm import tqdm
-from transformers import AutoModelForImageTextToText, AutoProcessor, GenerationConfig
+from transformers import AutoModel, AutoProcessor, GenerationConfig
 
 
 @torch.no_grad()
 def eval_mmmu(args):
     eval_args = EvalArgs.from_cli_args(args)
+    try:
+        from transformers import AutoModelForImageTextToText
+
+        model = AutoModelForImageTextToText.from_pretrained(
+            args.model_path,
+            torch_dtype="auto",
+            trust_remote_code=True,
+        )
+    except Exception as first_exception:
+        try:
+            model = AutoModel.from_pretrained(
+                args.model_path,
+                torch_dtype="auto",
+                trust_remote_code=True,
+                init_tts=False,
+            )
+        except Exception as second_exception:
+            raise RuntimeError(
+                f"Failed to load model: First attempt failed with {first_exception}, "
+                f"second attempt failed with {second_exception}"
+            ) from second_exception
 
-    model = AutoModelForImageTextToText.from_pretrained(
-        args.model_path,
-        torch_dtype="auto",
-        trust_remote_code=True,
-    )
     model = model.eval().cuda()
 
     processor = AutoProcessor.from_pretrained(
-        args.model_path, torch_dtype="auto", device_map="auto"
+        args.model_path, torch_dtype="auto", device_map="auto", trust_remote_code=True
     )
 
     samples = prepare_samples(eval_args)