Utility method for getting usage details from Response

[simonw]

Currently you can read usage numbers (from #610) like this:

response.input_tokens
response.output_tokens
response.token_details

But.. this only works if the response has been completed. It should work more like these methods:

llm/llm/models.py

Lines 350 to 355 in 02852fe

	def duration_ms(self) -> int:
	self._force()
	return int(((self._end or 0) - (self._start or 0)) * 1000)
	def datetime_utc(self) -> str:
	self._force()

And in the AsyncResponse:

llm/llm/models.py

Lines 442 to 448 in 02852fe

	async def duration_ms(self) -> int:
	await self._force()
	return int(((self._end or 0) - (self._start or 0)) * 1000)
	async def datetime_utc(self) -> str:
	await self._force()
	return self._start_utcnow.isoformat() if self._start_utcnow else ""

[simonw]

I think it's response.usage() and await response.usage() which call self._force() and then return the usage information as a dataclass with properties .input and .output and .details to match this set_usage() method:

llm/llm/models.py

Lines 215 to 224 in 02852fe

	def set_usage(
	self,
	*,
	input: Optional[int] = None,
	output: Optional[int] = None,
	details: Optional[dict] = None,
	):
	self.input_tokens = input
	self.output_tokens = output
	self.token_details = details

[simonw]

This will be documented in:

• Python API documentation for Response objects #643