phoenix-oss/llama-stack-mirror
0
0
Fork 1
mirror of https://github.com/meta-llama/llama-stack.git synced 2025-12-23 18:03:56 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

chore: add mypy inference fp8_impls

Browse source 
Signed-off-by: Mustafa Elbehery <melbeher@redhat.com>
This commit is contained in:
Mustafa Elbehery 2025-07-09 00:22:45 +02:00
parent d880c2df0e
commit 1c08a1cae9
7 changed files with 38 additions and 25 deletions
Download patch file Download diff file Expand all files Collapse all files

9
llama_stack/providers/remote/inference/vllm/vllm.py
View file

@ -168,7 +168,7 @@ def _process_vllm_chat_completion_end_of_stream(
last_chunk_content: str | None,
current_event_type: ChatCompletionResponseEventType,
tool_call_bufs: dict[str, UnparseableToolCall] | None = None,
) -> list[OpenAIChatCompletionChunk]:
) -> list[ChatCompletionResponseStreamChunk]:
chunks = []
if finish_reason is not None:
@ -247,9 +247,10 @@ async def _process_vllm_chat_completion_stream_response(
if choice.delta.tool_calls:
for delta_tool_call in choice.delta.tool_calls:
tool_call = convert_tool_call(delta_tool_call)
if delta_tool_call.index not in tool_call_bufs:
tool_call_bufs[delta_tool_call.index] = UnparseableToolCall()
tool_call_buf = tool_call_bufs[delta_tool_call.index]
index_str = str(delta_tool_call.index)
if index_str not in tool_call_bufs:
tool_call_bufs[index_str] = UnparseableToolCall()
tool_call_buf = tool_call_bufs[index_str]
tool_call_buf.tool_name += str(tool_call.tool_name)
tool_call_buf.call_id += tool_call.call_id
tool_call_buf.arguments += (