| 1 | # SPDX-License-Identifier: Apache-2.0 |
| 2 | # SPDX-FileCopyrightText: Copyright contributors to the vLLM project |
| 3 | from argparse import Namespace |
| 4 | from typing import TYPE_CHECKING |
| 5 | |
| 6 | from fastapi import FastAPI |
| 7 | |
| 8 | from vllm.engine.protocol import EngineClient |
| 9 | from vllm.logger import init_logger |
| 10 | from vllm.tasks import SupportedTask |
| 11 | |
| 12 | if TYPE_CHECKING: |
| 13 | from starlette.datastructures import State |
| 14 | |
| 15 | from vllm.entrypoints.serve.utils.request_logger import RequestLogger |
| 16 | else: |
| 17 | RequestLogger = object |
| 18 | |
| 19 | logger = init_logger(__name__) |
| 20 | |
| 21 | |
| 22 | def init_render_state( |
| 23 | state: "State", |
| 24 | request_logger: RequestLogger | None, |
| 25 | ): |
| 26 | from .derender.serving import ServingDerender |
| 27 | from .render.serving import ServingRender |
| 28 | |
| 29 | state.serving_render = ServingRender( |
| 30 | state.openai_serving_models, |
| 31 | state.online_renderer, |
| 32 | request_logger=request_logger, |
| 33 | tool_server=state.tool_server, |
| 34 | ) |
| 35 | |
| 36 | state.serving_derender = ServingDerender( |
| 37 | state.openai_serving_models, |
| 38 | state.online_derenderer, |
| 39 | request_logger=request_logger, |
| 40 | ) |
| 41 | |
| 42 | |
| 43 | def init_scale_out_state( |
| 44 | state: "State", |
| 45 | args: "Namespace", |
| 46 | engine_client: "EngineClient", |
| 47 | request_logger: RequestLogger | None, |
| 48 | ): |
| 49 | init_render_state(state, request_logger) |
| 50 | |
| 51 | from vllm.entrypoints.scale_out.token_in_token_out.serving import ServingTokens |
| 52 | |
| 53 | state.serving_tokens = ServingTokens( |
| 54 | engine_client, |
| 55 | state.openai_serving_models, |
| 56 | state.online_renderer, |
| 57 | request_logger=request_logger, |
| 58 | return_tokens_as_token_ids=args.return_tokens_as_token_ids, |
| 59 | enable_prompt_tokens_details=args.enable_prompt_tokens_details, |
| 60 | enable_log_outputs=args.enable_log_outputs, |
| 61 | force_no_detokenize=args.tokens_only, |
| 62 | ) |
| 63 | |
| 64 | |
| 65 | def register_scale_out_api_routers( |
| 66 | app: FastAPI, |
| 67 | supported_tasks: tuple["SupportedTask", ...], |
| 68 | ): |
| 69 | args = getattr(app.state, "args", None) |
| 70 | # `vllm launch render` and `vllm serve --tokens-only` are dedicated to |
| 71 | # serving these endpoints, so they always register them regardless of |
| 72 | # `--enable-scale-out`. |
| 73 | dedicated_mode = "render" in supported_tasks or getattr(args, "tokens_only", False) |
| 74 | enabled = dedicated_mode or getattr(args, "enable_scale_out", False) |
| 75 | if not enabled: |
| 76 | logger.info( |
| 77 | "Scale-out endpoints are disabled. Set --enable-scale-out to enable them." |
| 78 | ) |
| 79 | return |
| 80 | |
| 81 | from .render.api_router import router as render_render |
| 82 | |
| 83 | app.include_router(render_render) |
| 84 | |
| 85 | from .derender.api_router import router as derender_render |
| 86 | |
| 87 | app.include_router(derender_render) |
| 88 | |
| 89 | if "generate" in supported_tasks: |
| 90 | from .token_in_token_out.api_router import ( |
| 91 | attach_router as attach_disagg_router, |
| 92 | ) |
| 93 | |
| 94 | attach_disagg_router(app) |