feat(frontend): session id plumbing into requests (#48048)

Signed-off-by: Karen Chung <karenc@nvidia.com>
This commit is contained in:
Karen Chung
2026-08-03 22:18:58 +00:00
committed by GitHub
parent e578de311c
commit f57123aa2d
39 changed files with 438 additions and 4 deletions
+2
View File
@@ -145,6 +145,8 @@ class EngineCoreRequest(
# KV-transfer request is rejected on the D node before engine admission.
abort_immediately: bool = False
session_id: str | None = None
@property
def params(self) -> SamplingParams | PoolingParams:
"""Return the processed params (sampling or pooling)."""
+8
View File
@@ -294,6 +294,7 @@ class AsyncLLM(EngineClient):
trace_headers: Mapping[str, str] | None = None,
priority: int = 0,
data_parallel_rank: int | None = None,
session_id: str | None = None,
prompt_text: str | None = None,
reasoning_ended: bool | None = None,
reasoning_parser_kwargs: dict[str, Any] | None = None,
@@ -331,6 +332,7 @@ class AsyncLLM(EngineClient):
trace_headers,
priority,
data_parallel_rank,
session_id,
)
# Convert Input --> Request.
@@ -362,6 +364,7 @@ class AsyncLLM(EngineClient):
trace_headers=trace_headers,
priority=priority,
data_parallel_rank=data_parallel_rank,
session_id=session_id,
)
else:
# Raw prompts require tokenization and possibly multimodal
@@ -377,6 +380,7 @@ class AsyncLLM(EngineClient):
trace_headers=trace_headers,
priority=priority,
data_parallel_rank=data_parallel_rank,
session_id=session_id,
)
prompt_text, _, _ = extract_prompt_components(self.model_config, prompt)
@@ -445,6 +449,7 @@ class AsyncLLM(EngineClient):
trace_headers: Mapping[str, str] | None = None,
priority: int = 0,
data_parallel_rank: int | None = None,
session_id: str | None = None,
) -> RequestOutputCollector:
self._validate_streaming_input_sampling_params(sampling_params)
@@ -456,6 +461,7 @@ class AsyncLLM(EngineClient):
trace_headers=trace_headers,
priority=priority,
data_parallel_rank=data_parallel_rank,
session_id=session_id,
)
if not sampling_params.skip_clone:
@@ -556,6 +562,7 @@ class AsyncLLM(EngineClient):
trace_headers: Mapping[str, str] | None = None,
priority: int = 0,
data_parallel_rank: int | None = None,
session_id: str | None = None,
reasoning_ended: bool | None = None,
reasoning_parser_kwargs: dict[str, Any] | None = None,
) -> AsyncGenerator[RequestOutput, None]:
@@ -585,6 +592,7 @@ class AsyncLLM(EngineClient):
trace_headers=trace_headers,
priority=priority,
data_parallel_rank=data_parallel_rank,
session_id=session_id,
prompt_text=prompt_text,
reasoning_ended=reasoning_ended,
reasoning_parser_kwargs=reasoning_parser_kwargs,
+2
View File
@@ -261,6 +261,7 @@ class InputProcessor:
priority: int = 0,
data_parallel_rank: int | None = None,
resumable: bool = False,
session_id: str | None = None,
) -> EngineCoreRequest:
self._validate_params(params, supported_tasks)
self._validate_lora(lora_request)
@@ -391,6 +392,7 @@ class InputProcessor:
data_parallel_rank=data_parallel_rank,
trace_headers=trace_headers,
resumable=resumable,
session_id=session_id,
)
def _validate_prompt_len(
+2
View File
@@ -225,6 +225,7 @@ class LLMEngine:
tokenization_kwargs: dict[str, Any] | None = None,
trace_headers: Mapping[str, str] | None = None,
priority: int = 0,
session_id: str | None = None,
prompt_text: str | None = None,
) -> str:
# Validate the request_id type.
@@ -257,6 +258,7 @@ class LLMEngine:
tokenization_kwargs=tokenization_kwargs,
trace_headers=trace_headers,
priority=priority,
session_id=session_id,
)
prompt_text, _, _ = extract_prompt_components(self.model_config, prompt)