[video ingestion]Enable cosmos3#135
Closed
nv-liuw wants to merge 2 commits into
Closed
Conversation
Serve only the Cosmos3-Nano Reasoner (VLM) tower; the diffusion generator tower is never loaded (~8B-tier footprint). - serve.py: for "cosmos3" model names, add --hf-overrides Cosmos3ReasonerForConditionalGeneration (+ --mm-encoder-tp-mode data --async-scheduling) and skip the Qwen-only --mm-processor-kwargs; honor a VLLM_BIN env var so a dedicated vllm-cosmos3 venv can serve. - Dockerfile.cosmos: two-venv image (pipeline transformers 4.57 + an isolated cosmos vLLM server venv) because vllm-cosmos3 pulls transformers 5.x; apt installs ninja-build (required for vllm-cosmos3 kernel JIT at engine init). - osmo start_vllm.sh: auto-point VLLM_BIN at /opt/cosmos-venv when present (no-op on the default Qwen image). - configs: ingestion/batch_ingestion/retrieval point vlm_model/llm_model at nvidia/Cosmos3-Nano. - raise max_new_tokens on entity-extraction, dedup-merge, and caption calls so a reasoning model's output is not truncated. - docs: Model Backends gets a "Cosmos3-Nano (Reasoner)" section. (cherry picked from commit 9b9bddd)
…P compat - Dockerfile.cosmos: collapse to a single venv (vllm-cosmos3 + pipeline share one env); drop the second venv, VLLM_BIN, and the start_vllm.sh auto-detect. Keep ninja-build (vllm-cosmos3 needs it for kernel JIT). Image ~27GB (was ~38). - visual_extractor / search_frames: SigLIP get_image_features/get_text_features return a ModelOutput on transformers 5.x -> use .pooler_output; no-op on 4.x. - serve.py: drop VLLM_BIN (single venv: PATH `vllm` is the cosmos one). - configs: revert ingestion/batch/retrieval defaults to Qwen3-VL and document how to switch to nvidia/Cosmos3-Nano (zero impact for existing users). (cherry picked from commit 2503f54)
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.