Upload folder using huggingface_hub

modeling_internvl_chat.py

@@ -26,7 +26,7 @@ logger = logging.get_logger(__name__)
 class InternVLChatModel(PreTrainedModel):
     config_class = InternVLChatConfig
     main_input_name = 'pixel_values'
-    _no_split_modules = ['InternVisionEncoderLayer', 'LlamaDecoderLayer']
+    _no_split_modules = ['InternVisionEncoderLayer', 'LlamaDecoderLayer', 'InternLM2DecoderLayer']
 
     def __init__(self, config: InternVLChatConfig, vision_model=None, language_model=None):
         super().__init__(config)
@@ -237,10 +237,6 @@ class InternVLChatModel(PreTrainedModel):
             raise NotImplementedError
         img_context_token_id = tokenizer.convert_tokens_to_ids(IMG_CONTEXT_TOKEN)
         self.img_context_token_id = img_context_token_id
-        if tokenizer.convert_tokens_to_ids('<|im_end|>') != 0:
-            eos_token_id = tokenizer.convert_tokens_to_ids('<|im_end|>')  # 92542, InternLM2
-        else:
-            eos_token_id = tokenizer.eos_token_id
 
         from .conversation import get_conv_template
 
@@ -259,6 +255,7 @@ class InternVLChatModel(PreTrainedModel):
         model_inputs = tokenizer(queries, return_tensors='pt', padding=True)
         input_ids = model_inputs['input_ids'].cuda()
         attention_mask = model_inputs['attention_mask'].cuda()
+        eos_token_id = tokenizer.convert_tokens_to_ids(template.sep)
         generation_config['eos_token_id'] = eos_token_id
 
         generation_output = self.generate(
@@ -268,7 +265,7 @@ class InternVLChatModel(PreTrainedModel):
             **generation_config
         )
         responses = tokenizer.batch_decode(generation_output, skip_special_tokens=True)
-        responses = [response.split('<|im_end|>')[0].strip() for response in responses]  # for InternLM2
+        responses = [response.split(template.sep)[0].strip() for response in responses]
         return responses
 
     def chat(self, tokenizer, pixel_values, question, generation_config, history=None, return_history=False,
@@ -276,10 +273,6 @@ class InternVLChatModel(PreTrainedModel):
 
         img_context_token_id = tokenizer.convert_tokens_to_ids(IMG_CONTEXT_TOKEN)
         self.img_context_token_id = img_context_token_id
-        if tokenizer.convert_tokens_to_ids('<|im_end|>') != 0:
-            eos_token_id = tokenizer.convert_tokens_to_ids('<|im_end|>')  # 92542, InternLM2
-        else:
-            eos_token_id = tokenizer.eos_token_id
 
         from .conversation import get_conv_template
 
@@ -300,7 +293,9 @@ class InternVLChatModel(PreTrainedModel):
         model_inputs = tokenizer(query, return_tensors='pt')
         input_ids = model_inputs['input_ids'].cuda()
         attention_mask = model_inputs['attention_mask'].cuda()
+        eos_token_id = tokenizer.convert_tokens_to_ids(template.sep)
         generation_config['eos_token_id'] = eos_token_id
+
         generation_output = self.generate(
             pixel_values=pixel_values,
             input_ids=input_ids,
@@ -308,7 +303,7 @@ class InternVLChatModel(PreTrainedModel):
             **generation_config
         )
         response = tokenizer.batch_decode(generation_output, skip_special_tokens=True)[0]
-        response = response.split('<|im_end|>')[0].strip()  # for InternLM2
+        response = response.split(template.sep)[0].strip()
         history.append((question, response))
         if return_history:
             return response, history