llava-hf

/

LLaVA-NeXT-Video-7B-hf

like 50

Follow

Llava Hugging Face 121

Dear All,

When running this cell:
generate_kwargs = {"max_new_tokens": 100, "do_sample": True, "top_p": 0.9}

output = model.generate(**inputs, **generate_kwargs)
generated_text = processor.batch_decode(output, skip_special_tokens=True)

I got this error:

Expanding inputs for image.video tokens in LLaVa-NeXT-Video should be done in processing. Please add patch_size and vision_feature_select_strategy to the model's processing config or set directly with processor.patch_size = {{patch_size}} and processor.vision_feature_select_strategy = {{vision_feature_select_strategy}}`. Using processors without these attributes in the config is deprecated and will throw an error in v4.47.

RuntimeError Traceback (most recent call last)

in <cell line: 3>()
1 generate_kwargs = {"max_new_tokens": 100, "do_sample": True, "top_p": 0.9}
2
----> 3 output = model.generate(**inputs, **generate_kwargs)
4 generated_text = processor.batch_decode(output, skip_special_tokens=True)

22 frames

/usr/local/lib/python3.10/dist-packages/transformers/models/llama/modeling_llama.py in forward(self, hidden_states, attention_mask, position_ids, past_key_value, output_attentions, use_cache, cache_position, position_embeddings, **kwargs)
600 is_causal = True if causal_mask is None and q_len > 1 else False
601
--> 602 attn_output = torch.nn.functional.scaled_dot_product_attention(
603 query_states,
604 key_states,

RuntimeError: The expanded size of the tensor (1172) must match the existing size (21) at non-singleton dimension 3. Target sizes: [2, 32, 1, 1172]. Tensor sizes: [2, 1, 1, 21]