Repository navigation

All pull requests

Labels (9)Milestones (0)New pull request

fix: export adapter for stage-two training
#21dajiaohuang opened Oct 25, 2026
fix: honor configured optimizer warmup ratio
#20dajiaohuang opened Oct 25, 2026
fix: use configured AdamW epsilon
#19dajiaohuang opened Oct 25, 2026
fix: preserve real end-of-turn tokens in attention masks
#18dajiaohuang opened Oct 25, 2026
fix: pass planned task type to graph action agent
#17dajiaohuang opened Oct 25, 2026
fix: cast graph projector during training setup
#16dajiaohuang opened Oct 25, 2026
fix: align attention mask with model device
#13dajiaohuang opened Oct 25, 2026
fix: preserve all heterogeneous graph features
#12dajiaohuang opened Oct 25, 2026
fix: align scaffold graph data by node id
#11dajiaohuang opened Oct 25, 2026
fix: preserve literal graph markers in user prompts
#10dajiaohuang opened Oct 25, 2026
fix: make inference launcher start from repository root
#9dajiaohuang opened Oct 25, 2026