[Mini-VLLM] Remove LoRA, plugins, speech-to-text and scale-out surfaces
Production runs one static model behind the OpenAI-compatible API. Every adapter path and non-OpenAI HTTP surface is unreachable there. - Delete the LoRA stack: vllm/lora, LoRARequest on the engine wire, the scheduler, worker, cudagraph and MoE plumbing, LoRA metrics, the lora CLI group, and the LoRA tests. - Drop the plugin loader in vllm/plugins. Platform resolution is now CUDA-only and the general/stat-logger plugin hooks are gone. - Delete the speech-to-text and realtime entrypoints, the scale-out render routes, and vllm/config/speech_to_text.py. GenerationTask is generate-only. - Cut DecodeBenchConnector and the disagg-prefill doc and image. The Multi, Offloading and Example connectors stay; production uses the CPU KV offload tier. - Delete the ROCm-only fused KDA chunk kernel (2346 lines) with its CMake arch gate, declarations and op registrations. The CUDA build never defined the guard and no Python caller existed. - Delete the orphan Mistral and Pixtral chat fixtures and examples/. - Fix _check_model: removing the LoRA early return made every request fail with "The model ... does not exist. Valid aliases: ...". Verified on an RTX 3090: import sweep reports 1242 modules with 7 known optional-dependency failures; a live server answers chat, completions, responses, tokenize and detokenize with zero log errors and 404 for the removed surfaces; collection yields 8558 tests with the 3 pre-existing errors; ruff adds no new finding; the incremental _C build succeeds. Co-authored-by: Hinata <hinata@idihore.id> Signed-off-by: MOVZX <movzx@yahoo.com>
This commit is contained in:
@@ -10,7 +10,6 @@ from typing import TYPE_CHECKING, Any, cast
|
||||
import numpy as np
|
||||
import torch
|
||||
|
||||
from vllm.lora.request import LoRARequest
|
||||
from vllm.outputs import (
|
||||
STREAM_FINISHED,
|
||||
CompletionOutput,
|
||||
@@ -36,10 +35,8 @@ from vllm.v1.engine.logprobs import LogprobsProcessor
|
||||
from vllm.v1.engine.parallel_sampling import ParentRequest
|
||||
from vllm.v1.metrics.stats import (
|
||||
IterationStats,
|
||||
LoRARequestStates,
|
||||
RequestSpecDecodeMetrics,
|
||||
RequestStateStats,
|
||||
SchedulerStats,
|
||||
)
|
||||
from vllm.v1.outputs import SamplingMaskLists
|
||||
|
||||
@@ -187,7 +184,6 @@ class RequestState:
|
||||
external_req_id: str,
|
||||
parent_req: ParentRequest | None,
|
||||
request_index: int,
|
||||
lora_request: LoRARequest | None,
|
||||
output_kind: RequestOutputKind,
|
||||
prompt: str | None,
|
||||
prompt_token_ids: list[int] | None,
|
||||
@@ -209,8 +205,6 @@ class RequestState:
|
||||
self.external_req_id = external_req_id
|
||||
self.parent_req = parent_req
|
||||
self.request_index = request_index
|
||||
self.lora_request = lora_request
|
||||
self.lora_name = lora_request.lora_name if lora_request is not None else None
|
||||
self.output_kind = output_kind
|
||||
self.prompt = prompt
|
||||
self.prompt_token_ids = prompt_token_ids
|
||||
@@ -328,7 +322,6 @@ class RequestState:
|
||||
external_req_id=request.external_req_id,
|
||||
parent_req=parent_req,
|
||||
request_index=request_index,
|
||||
lora_request=request.lora_request,
|
||||
output_kind=output_kind,
|
||||
prompt=prompt,
|
||||
prompt_token_ids=request.prompt_token_ids,
|
||||
@@ -458,7 +451,6 @@ class RequestState:
|
||||
|
||||
return RequestOutput(
|
||||
request_id=external_req_id, # request_id is what was provided externally
|
||||
lora_request=self.lora_request,
|
||||
prompt=self.prompt,
|
||||
prompt_token_ids=prompt_token_ids,
|
||||
prompt_logprobs=prompt_logprobs,
|
||||
@@ -566,7 +558,6 @@ class OutputProcessor:
|
||||
self.request_states: dict[str, RequestState] = {}
|
||||
self.parent_requests: dict[str, ParentRequest] = {}
|
||||
self.external_req_ids: defaultdict[str, list[str]] = defaultdict(list)
|
||||
self.lora_states = LoRARequestStates(log_stats)
|
||||
self.tracing_enabled = tracing_enabled
|
||||
self.admission_stats = admission_stats
|
||||
|
||||
@@ -632,7 +623,6 @@ class OutputProcessor:
|
||||
for request_id in internal_req_ids:
|
||||
req_state = self.request_states.pop(request_id, None)
|
||||
if req_state is not None:
|
||||
self.lora_states.request_finished(request_id, req_state.lora_name)
|
||||
request_ids_to_abort.append(request_id)
|
||||
# Produce final abort output.
|
||||
if req_state.queue is not None and (
|
||||
@@ -923,9 +913,6 @@ class OutputProcessor:
|
||||
if self.admission_stats is not None:
|
||||
self.admission_stats.set_num_requests(self.get_num_unfinished_requests())
|
||||
|
||||
def update_scheduler_stats(self, scheduler_stats: SchedulerStats | None):
|
||||
self.lora_states.update_scheduler_stats(scheduler_stats)
|
||||
|
||||
def do_tracing(
|
||||
self,
|
||||
engine_core_output: EngineCoreOutput,
|
||||
@@ -1005,8 +992,6 @@ class OutputProcessor:
|
||||
engine_core_timestamp,
|
||||
req_state.is_prefilling,
|
||||
req_state.stats,
|
||||
self.lora_states,
|
||||
req_state.lora_name,
|
||||
)
|
||||
|
||||
def _update_stats_from_finished(
|
||||
@@ -1028,7 +1013,6 @@ class OutputProcessor:
|
||||
req_stats=req_state.stats,
|
||||
num_cached_tokens=req_state.num_cached_tokens,
|
||||
)
|
||||
self.lora_states.request_finished(req_state.request_id, req_state.lora_name)
|
||||
|
||||
ParentRequest.observe_finished_request(
|
||||
req_state.parent_req, iteration_stats, req_state.stats.num_generation_tokens
|
||||
|
||||
Reference in New Issue
Block a user