Consolidates REST-based LLMs.

[copybara-service[bot]]

Consolidates REST-based LLMs.

This CL adds lf.llms.REST and consolidates Anthropic/Groq models based on it. The motivation are two-folds: a) We need to better iterate the shared code among different REST-based models; b) Allow users to quickly develop new REST-based clients in a few lines of code, without worrying about concurrency and error control.

lf.llms.REST can be used in two ways:

1) Composition:

lf.llms.REST(
  api_endpoint='...', # URL to call.
  headers=dict(...)   # headers for model choice and authentication.
  request=lambda prompt, sampling_options: dict(...),
  result=lambda json: lf.llms.LMSamplingResult(...),
  max_concurrency=8,  # Concurrency control.
) 

2) Subclassing:

class MyLLM(lf.llms.REST):
  model: str

  @property  
  def headers(self):
    return {
       'model': self.model,
    }
  def request(self, prompt, sampling_options) -> dict[str, Any]:
    ...
  def result(self, json) -> lf.LMSamplingResult:
    ...