gopidesupavan commented on code in PR #68372:
URL: https://github.com/apache/airflow/pull/68372#discussion_r3416664704
##########
providers/common/ai/src/airflow/providers/common/ai/durable/caching_model.py:
##########
@@ -67,15 +73,33 @@ async def request(
) -> ModelResponse:
step = self.counter.next_step()
key = f"model_step_{step}"
+ fingerprint = fingerprint_model_request(
Review Comment:
can this fingerprint be computed from the model prepare request rather than
aw `model_settings` / `model_request_parameters` passed into `request()`. why i
am thinking is most of the models calls the prepare_request() before sending to
the provider, that merges model-level `self.settings`, applies
model/profile-specific. eg thinking and some filters native tools, so the
current placement, a retry can replay a cached response when the raw Agent
inputs are unchanged but the actual provider request would differ after
preparation.
may be we can use something like?
self.wrapped.prepare_request(model_settings, model_request_parameters)
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
