Remove unnecessary lines since use batched visuals now in llava · EvolvingLMMs-Lab/lmms-eval@12f4806 (original) (raw)
`@@ -358,18 +358,6 @@ def _collate(x):
`
358
358
`prompt_question = conv.get_prompt()
`
359
359
`question_input.append(prompt_question)
`
360
360
``
361
``
`-
The above for loop has bugs. When there is no visuals, e.g. pure text,
`
362
``
`-
there will be no for loop execute resulting in an empty question_input (because no visuals)
`
363
``
`-
Scenario 1 won't even be execute
`
364
``
`-
if len(flattened_visuals) == 0:
`
365
``
`-
for context in contexts:
`
366
``
`-
question = context
`
367
``
`-
conv = conv_templates[self.conv_template].copy()
`
368
``
`-
conv.append_message(conv.roles[0], question)
`
369
``
`-
conv.append_message(conv.roles[1], None)
`
370
``
`-
prompt_question = conv.get_prompt()
`
371
``
`-
question_input.append(prompt_question)
`
372
``
-
373
361
`# input_ids = tokenizer_image_token(prompt, self.tokenizer, IMAGE_TOKEN_INDEX, return_tensors="pt").unsqueeze(0).to(self.device)
`
374
362
`# preconfigure gen_kwargs with defaults
`
375
363
`gen_kwargs["image_sizes"] = [flattened_visuals[idx].size for idx in range(len(flattened_visuals))]
`