Pepijn 3d7e60cee4 feat(annotate): parallel_servers spawns N independent vllm replicas ...

Adds --vlm.parallel_servers=N. Spawns N independent vllm processes
(each pinned to GPU i via CUDA_VISIBLE_DEVICES, listening on
serve_port+i) and round-robins requests across them. Sidesteps DP/TP
NCCL setup failures on nodes with restricted P2P/SHM.

Default serve_command for parallel mode: vllm serve <model_id>
--tensor-parallel-size 1 --max-model-len 32768 --uvicorn-log-level
warning. Override via --vlm.serve_command (use {port} placeholder).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

2026-04-30 18:48:35 +02:00

..

modules

feat(annotate): Module 1 samples image frames at fps rate

2026-04-30 18:48:35 +02:00

prompts

feat(annotate): Module 1 sees the whole episode as one video block

2026-04-30 18:48:33 +02:00

__init__.py

feat: language annotation pipeline (PR 2/3)

2026-04-30 18:48:33 +02:00

config.py

feat(annotate): parallel_servers spawns N independent vllm replicas

2026-04-30 18:48:35 +02:00

executor.py

feat(annotate): per-episode progress logs in executor

2026-04-30 18:48:35 +02:00

frames.py

feat(annotate): video_url block for openai backend

2026-04-30 18:48:34 +02:00

reader.py

feat: language annotation pipeline (PR 2/3)

2026-04-30 18:48:33 +02:00

staging.py

feat: language annotation pipeline (PR 2/3)

2026-04-30 18:48:33 +02:00

validator.py

feat: language annotation pipeline (PR 2/3)

2026-04-30 18:48:33 +02:00

vlm_client.py

feat(annotate): parallel_servers spawns N independent vllm replicas

2026-04-30 18:48:35 +02:00

writer.py

feat: language annotation pipeline (PR 2/3)

2026-04-30 18:48:33 +02:00