phoenix/litellm-mirror
0
0
Fork 1
mirror of https://github.com/BerriAI/litellm.git synced 2025-04-25 18:54:30 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Merge pull request #3905 from giritatavarty-8451/litellm_triton_chatcompletion_support

Browse source 
Litellm triton chatcompletion support - Resubmit of #3895
This commit is contained in:
Ishaan Jaff 2024-07-23 10:30:26 -07:00 • committed by GitHub
commit 1355932bf4
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
3 changed files with 282 additions and 22 deletions
Download patch file Download diff file Expand all files Collapse all files

43
litellm/utils.py
View file

@ -9097,6 +9097,42 @@ class CustomStreamWrapper:
except Exception as e:
raise e
def handle_triton_stream(self, chunk):
try:
if isinstance(chunk, dict):
parsed_response = chunk
elif isinstance(chunk, (str, bytes)):
if isinstance(chunk, bytes):
chunk = chunk.decode("utf-8")
if "text_output" in chunk:
response = chunk.replace("data: ", "").strip()
parsed_response = json.loads(response)
else:
return {
"text": "",
"is_finished": False,
"prompt_tokens": 0,
"completion_tokens": 0,
}
else:
print_verbose(f"chunk: {chunk} (Type: {type(chunk)})")
raise ValueError(
f"Unable to parse response. Original response: {chunk}"
)
text = parsed_response.get("text_output", "")
finish_reason = parsed_response.get("stop_reason")
is_finished = parsed_response.get("is_finished", False)
return {
"text": text,
"is_finished": is_finished,
"finish_reason": finish_reason,
"prompt_tokens": parsed_response.get("input_token_count", 0),
"completion_tokens": parsed_response.get("generated_token_count", 0),
}
return {"text": "", "is_finished": False}
except Exception as e:
raise e
def handle_clarifai_completion_chunk(self, chunk):
try:
if isinstance(chunk, dict):
@ -9516,6 +9552,12 @@ class CustomStreamWrapper:
completion_obj["content"] = response_obj["text"]
if response_obj["is_finished"]:
self.received_finish_reason = response_obj["finish_reason"]
elif self.custom_llm_provider == "triton":
response_obj = self.handle_triton_stream(chunk)
completion_obj["content"] = response_obj["text"]
print_verbose(f"completion obj content: {completion_obj['content']}")
if response_obj["is_finished"]:
self.received_finish_reason = response_obj["finish_reason"]
elif self.custom_llm_provider == "text-completion-openai":
response_obj = self.handle_openai_text_completion_chunk(chunk)
completion_obj["content"] = response_obj["text"]
@ -10071,6 +10113,7 @@ class CustomStreamWrapper:
or self.custom_llm_provider == "predibase"
or self.custom_llm_provider == "databricks"
or self.custom_llm_provider == "bedrock"
or self.custom_llm_provider == "triton"
or self.custom_llm_provider == "watsonx"
or self.custom_llm_provider in litellm.openai_compatible_endpoints
):