Add SFT+GRPO interpretability report and sync COMMAND_SPEC - #4
Merged
Conversation
Sync both COMMAND_SPEC.md and COMMAND_SPEC_zh.md with the canonical command specification from the tool schema source. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Signed-off-by: zzzhang20 <zzzhang20@iflytek.com>
Post-hoc analysis of why the two-stage SFT+GRPO training works: SFT does the main capability acquisition (training reward saturates at ~0.999), while GRPO performs targeted long-tail refinement (canonicalization and hallucinated-slot suppression, fix:break ~6:1 on the paired test set). Includes SwanLab reward curves and reproducibility notes. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Signed-off-by: zzzhang20 <zzzhang20@iflytek.com>
zzzhangstat
force-pushed
the
docs/update-readme
branch
from
June 29, 2026 08:06
456569c to
c9c6cb9
Compare
This was referenced Aug 23, 2026
dongjiang1989
pushed a commit
that referenced
this pull request
Aug 24, 2026
* docs: add customized HER Hack-Astron issue template Signed-off-by: FenjuFu <fufenjupku@gmail.com> * feat: add Domux HER Hack-Astron case framework Signed-off-by: FenjuFu <92919259+FenjuFu@users.noreply.github.com> * docs: remove internal campaign metrics Signed-off-by: FenjuFu <92919259+FenjuFu@users.noreply.github.com> --------- Signed-off-by: FenjuFu <fufenjupku@gmail.com> Signed-off-by: FenjuFu <92919259+FenjuFu@users.noreply.github.com>
This was referenced Aug 29, 2026
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
What's changed
two-stage training works: SFT does the main capability acquisition
(training reward saturates at ~0.999), while GRPO performs targeted
long-tail refinement (canonicalization + hallucinated-slot suppression,
fix:break ≈ 6:1)
New files
GRPO_Interpretability_Report.md/_zh.mdassets/SwanLab-Chart_SlotAccuracy.png/SlotFormat.png