Will microsoft/Phi-3-vision-128k-instruct support transformers v4.51.3?
#71
by
Huaiyu666
- opened
When I run microsoft/Phi-3-vision-128k-instruct with transformers v4.51.3, I got the error: failed: AttributeError: 'DynamicCache' object has no attribute 'get_max_length'.
details:
"/root/.cache/huggingface/modules/transformers_modules/microsoft/Phi_3_vision_128k_instruct/_45209e90a4c4f7d16b2e9d48503c7f3e83623ed/modeling_phi3_v.py", line 1352, in prepare_inputs_for_generation
max_cache_length = past_key_values.get_max_length()
AttributeError: 'DynamicCache' object has no attribute 'get_max_length'. Did you mean: 'get_seq_length'?
Setting use_cache=False in the generation_args fixes the issue for me. It slows everything down, however.