phoenix-oss/llama-stack-mirror
0
0
Fork 1
mirror of https://github.com/meta-llama/llama-stack.git synced 2025-10-23 08:33:09 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Merge remote-tracking branch 'origin/main' into stores
Some checks failed
Installer CI / smoke-test-on-dev (push) Failing after 3s
Details
Installer CI / lint (push) Failing after 3s
Details

Browse source
This commit is contained in:
Ashwin Bharambe 2025-10-13 11:07:11 -07:00
commit b72154ce5e
1161 changed files with 609896 additions and 42960 deletions
Download patch file Download diff file Expand all files Collapse all files

40
llama_stack/providers/inline/batches/reference/batches.py
View file

@ -22,7 +22,10 @@ from llama_stack.apis.files import Files, OpenAIFilePurpose
from llama_stack.apis.inference import (
Inference,
OpenAIAssistantMessageParam,
OpenAIChatCompletionRequestWithExtraBody,
OpenAICompletionRequestWithExtraBody,
OpenAIDeveloperMessageParam,
OpenAIEmbeddingsRequestWithExtraBody,
OpenAIMessageParam,
OpenAISystemMessageParam,
OpenAIToolMessageParam,
@ -178,9 +181,9 @@ class ReferenceBatchesImpl(Batches):
# TODO: set expiration time for garbage collection
if endpoint not in ["/v1/chat/completions", "/v1/completions"]:
if endpoint not in ["/v1/chat/completions", "/v1/completions", "/v1/embeddings"]:
raise ValueError(
f"Invalid endpoint: {endpoint}. Supported values: /v1/chat/completions, /v1/completions. Code: invalid_value. Param: endpoint",
f"Invalid endpoint: {endpoint}. Supported values: /v1/chat/completions, /v1/completions, /v1/embeddings. Code: invalid_value. Param: endpoint",
)
if completion_window != "24h":
@ -425,18 +428,23 @@ class ReferenceBatchesImpl(Batches):
valid = False
if batch.endpoint == "/v1/chat/completions":
required_params = [
required_params: list[tuple[str, Any, str]] = [
("model", str, "a string"),
# messages is specific to /v1/chat/completions
# we could skip validating messages here and let inference fail. however,
# that would be a very expensive way to find out messages is wrong.
("messages", list, "an array"), # TODO: allow messages to be a string?
]
else: # /v1/completions
elif batch.endpoint == "/v1/completions":
required_params = [
("model", str, "a string"),
("prompt", str, "a string"), # TODO: allow prompt to be a list of strings??
]
else: # /v1/embeddings
required_params = [
("model", str, "a string"),
("input", (str, list), "a string or array of strings"),
]
for param, expected_type, type_string in required_params:
if param not in body:
@ -601,7 +609,8 @@ class ReferenceBatchesImpl(Batches):
# TODO(SECURITY): review body for security issues
if request.url == "/v1/chat/completions":
request.body["messages"] = [convert_to_openai_message_param(msg) for msg in request.body["messages"]]
chat_response = await self.inference_api.openai_chat_completion(**request.body)
chat_params = OpenAIChatCompletionRequestWithExtraBody(**request.body)
chat_response = await self.inference_api.openai_chat_completion(chat_params)
# this is for mypy, we don't allow streaming so we'll get the right type
assert hasattr(chat_response, "model_dump_json"), "Chat response must have model_dump_json method"
@ -614,8 +623,9 @@ class ReferenceBatchesImpl(Batches):
"body": chat_response.model_dump_json(),
},
}
else: # /v1/completions
completion_response = await self.inference_api.openai_completion(**request.body)
elif request.url == "/v1/completions":
completion_params = OpenAICompletionRequestWithExtraBody(**request.body)
completion_response = await self.inference_api.openai_completion(completion_params)
# this is for mypy, we don't allow streaming so we'll get the right type
assert hasattr(completion_response, "model_dump_json"), (
@ -630,6 +640,22 @@ class ReferenceBatchesImpl(Batches):
"body": completion_response.model_dump_json(),
},
}
else: # /v1/embeddings
embeddings_response = await self.inference_api.openai_embeddings(
OpenAIEmbeddingsRequestWithExtraBody(**request.body)
)
assert hasattr(embeddings_response, "model_dump_json"), (
"Embeddings response must have model_dump_json method"
)
return {
"id": request_id,
"custom_id": request.custom_id,
"response": {
"status_code": 200,
"request_id": request_id, # TODO: should this be different?
"body": embeddings_response.model_dump_json(),
},
}
except Exception as e:
logger.info(f"Error processing request {request.custom_id} in batch {batch_id}: {e}")
return {