microsoft

/

Phi-3.5-vision-instruct

like 651

Follow

Microsoft 7.88k

Extend eos_token_id with pad token because that's the token stored in processor.tokenizer.eos_token_id. That also simplifies the sample from the model card from

generate_ids = model.generate(**inputs, 
  eos_token_id=processor.tokenizer.eos_token_id, 
  **generation_args
)

to

generate_ids = model.generate(**inputs, 
  **generation_args
)