Merge branch 'main' into litellm_aioboto3_sagemaker

This commit is contained in:
Krish Dholakia 2024-02-14 21:46:58 -08:00 committed by GitHub
commit 57654f4533
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
79 changed files with 3440 additions and 253 deletions

View file

@ -738,6 +738,8 @@ class CallTypes(Enum):
text_completion = "text_completion"
image_generation = "image_generation"
aimage_generation = "aimage_generation"
moderation = "moderation"
amoderation = "amoderation"
# Logging function -> log the exact model details + what's being sent | Non-BlockingP
@ -2100,6 +2102,11 @@ def client(original_function):
or call_type == CallTypes.aimage_generation.value
):
messages = args[0] if len(args) > 0 else kwargs["prompt"]
elif (
call_type == CallTypes.moderation.value
or call_type == CallTypes.amoderation.value
):
messages = args[1] if len(args) > 1 else kwargs["input"]
elif (
call_type == CallTypes.atext_completion.value
or call_type == CallTypes.text_completion.value
@ -7692,6 +7699,7 @@ class CustomStreamWrapper:
self.special_tokens = ["<|assistant|>", "<|system|>", "<|user|>", "<s>", "</s>"]
self.holding_chunk = ""
self.complete_response = ""
self.response_uptil_now = ""
_model_info = (
self.logging_obj.model_call_details.get("litellm_params", {}).get(
"model_info", {}
@ -7703,6 +7711,7 @@ class CustomStreamWrapper:
} # returned as x-litellm-model-id response header in proxy
self.response_id = None
self.logging_loop = None
self.rules = Rules()
def __iter__(self):
return self
@ -8659,7 +8668,7 @@ class CustomStreamWrapper:
chunk = next(self.completion_stream)
if chunk is not None and chunk != b"":
print_verbose(f"PROCESSED CHUNK PRE CHUNK CREATOR: {chunk}")
response = self.chunk_creator(chunk=chunk)
response: Optional[ModelResponse] = self.chunk_creator(chunk=chunk)
print_verbose(f"PROCESSED CHUNK POST CHUNK CREATOR: {response}")
if response is None:
continue
@ -8667,7 +8676,12 @@ class CustomStreamWrapper:
threading.Thread(
target=self.run_success_logging_in_thread, args=(response,)
).start() # log response
self.response_uptil_now += (
response.choices[0].delta.get("content", "") or ""
)
self.rules.post_call_rules(
input=self.response_uptil_now, model=self.model
)
# RETURN RESULT
return response
except StopIteration:
@ -8705,7 +8719,9 @@ class CustomStreamWrapper:
# chunk_creator() does logging/stream chunk building. We need to let it know its being called in_async_func, so we don't double add chunks.
# __anext__ also calls async_success_handler, which does logging
print_verbose(f"PROCESSED ASYNC CHUNK PRE CHUNK CREATOR: {chunk}")
processed_chunk = self.chunk_creator(chunk=chunk)
processed_chunk: Optional[ModelResponse] = self.chunk_creator(
chunk=chunk
)
print_verbose(
f"PROCESSED ASYNC CHUNK POST CHUNK CREATOR: {processed_chunk}"
)
@ -8720,6 +8736,12 @@ class CustomStreamWrapper:
processed_chunk,
)
)
self.response_uptil_now += (
processed_chunk.choices[0].delta.get("content", "") or ""
)
self.rules.post_call_rules(
input=self.response_uptil_now, model=self.model
)
return processed_chunk
raise StopAsyncIteration
else: # temporary patch for non-aiohttp async calls
@ -8733,7 +8755,9 @@ class CustomStreamWrapper:
chunk = next(self.completion_stream)
if chunk is not None and chunk != b"":
print_verbose(f"PROCESSED CHUNK PRE CHUNK CREATOR: {chunk}")
processed_chunk = self.chunk_creator(chunk=chunk)
processed_chunk: Optional[ModelResponse] = self.chunk_creator(
chunk=chunk
)
print_verbose(
f"PROCESSED CHUNK POST CHUNK CREATOR: {processed_chunk}"
)
@ -8750,6 +8774,12 @@ class CustomStreamWrapper:
)
)
self.response_uptil_now += (
processed_chunk.choices[0].delta.get("content", "") or ""
)
self.rules.post_call_rules(
input=self.response_uptil_now, model=self.model
)
# RETURN RESULT
return processed_chunk
except StopAsyncIteration: