Repository navigation
data: DeepSeek-V4 thinking mode for regeneration and training - #1069
Draft
yushengsu-thu wants to merge 1 commit into
Draft
yushengsu-thu wants to merge 1 commit into
yushengsu-thu wants to merge 1 commit into
Conversation
Regeneration sends the thinking key SGLang reads for DeepSeek-V4; deepseek-v4-thinking renders reasoning in training.
yushengsu-thu
requested review from
FlamingoPg,
shuaills,
sleepcoo and
zyksir
as code owners
November 2, 2026 22:55
yushengsu-thu
marked this pull request as draft
November 3, 2026 00:57
yushengsu-thu
marked this pull request as ready for review
November 3, 2026 04:24
yushengsu-thu
marked this pull request as draft
November 3, 2026 04:25
This branch has not been deployed
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Makes DeepSeek-V4 thinking-mode data usable end to end.
scripts/regenerate_train_data.py:--reasoning saveand--reasoning disableonly sent Qwen'senable_thinking, but SGLang readschat_template_kwargs.thinkingfor DeepSeek-V4, sosavegot chat-mode replies with noreasoning_contentand every row errored. Both keys are now sent.specforge/data/template.py: addsdeepseek-v4-thinking, the existingdeepseek-v4template withenable_thinking=True, so training renders the reasoning.Tests: devbox CPU run of
tests/test_data/test_deepseek_v4_template.pyandtests/test_data/test_template_registry.py: 14 passed. No new test case. The regenerate change is in use on a live V4-Flash 0731 regeneration: 33,392 of 100,000 prompts processed, 29,967 written so far.