Allow llava_ov SFT to start from VLM_SAFETENSORS_PATH#45
Open
foreverlms wants to merge 1 commit into
Open
Conversation
Mirror launch_sft_videophy2_nano.sh: when VLM_SAFETENSORS_PATH is set, plumb it to model.config.policy.backbone.safetensors_path via a tail override so llava_ov Reasoner-alignment SFT can bootstrap from a merged Cosmos3 reasoner snapshot instead of the public Qwen3-VL-8B-Instruct backbone. Unset behavior is unchanged. Update training.md accordingly. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
foreverlms
force-pushed
the
maoshengl/update_llava_ov
branch
from
0312655 to
e533d67
Compare
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Mirror launch_sft_videophy2_nano.sh: when VLM_SAFETENSORS_PATH is set, plumb it to model.config.policy.backbone.safetensors_path via a tail override so llava_ov Reasoner-alignment SFT can bootstrap from a merged Cosmos3 reasoner snapshot instead of the public Qwen3-VL-8B-Instruct backbone. Unset behavior is unchanged. Update training.md accordingly.