Is there a way to control the temperature of the model?

#19
by zkdtckk - opened

The model performs really awesome, and I am trying to find a way to make it more repeatable. Any parameter to adjust like the temperature=0 in OpenAI GPT API?

Technology Innovation Institute org

Yup, the options are here: https://huggingface.co/docs/transformers/main_classes/text_generation#transformers.GenerationConfig
Eg. the repetition penalty might be useful.

You should be able to specify them when calling the pipeline

sequences = pipeline(
   "prompt goes here",
    max_length=200,
    do_sample=True,
    top_k=10,
    num_return_sequences=1,
    eos_token_id=tokenizer.eos_token_id,
    repetition_penalty=2.0
)
FalconLLM changed discussion status to closed

Sign up or log in to comment