Video-Text-to-Text
Safetensors
custom_code
ynhe commited on
Commit
095df6f
1 Parent(s): de18473

[fix] fix image ilen

Browse files
Files changed (1) hide show
  1. modeling_videochat2.py +3 -1
modeling_videochat2.py CHANGED
@@ -257,7 +257,7 @@ class InternVideo2_VideoChat2(BaseMLLM):
257
  return_history =False,
258
  generation_config={}
259
  ):
260
- ilen = media_tensor.shape[1]
261
 
262
  conversation = ""
263
  if instruction:
@@ -268,8 +268,10 @@ class InternVideo2_VideoChat2(BaseMLLM):
268
  )
269
 
270
  if media_type == 'image':
 
271
  conversation +=( "<img>" + IMG_TOKEN + "</img>")*ilen
272
  else:
 
273
  conversation += ("<vid>" + VID_TOKEN + "</vid>")*ilen
274
 
275
 
 
257
  return_history =False,
258
  generation_config={}
259
  ):
260
+
261
 
262
  conversation = ""
263
  if instruction:
 
268
  )
269
 
270
  if media_type == 'image':
271
+ ilen = media_tensor.shape[0]
272
  conversation +=( "<img>" + IMG_TOKEN + "</img>")*ilen
273
  else:
274
+ ilen = media_tensor.shape[1]
275
  conversation += ("<vid>" + VID_TOKEN + "</vid>")*ilen
276
 
277