Repository navigation

Fix continual DPO trainer lifecycle, explicit eval/logging, and DeepSpeed guidance - #210

Draft

EMZEDI with Copilot wants to merge 16 commits into

mainfrom

copilot/fix-continual-dpo-training-pipeline

Commits on Jul 9, 2026

Initial plan
Copilot
authored
88c8af4
Fix continual DPO trainer lifecycle
Copilot
authored
0835077
Address DPO review fixes
Copilot
authored
57fbb43
Polish continual DPO edge cases
Copilot
authored
87f8a74
Tidy DPO script imports
Copilot
authored
c172f50
Clarify DPO eval dataloaders
Copilot
authored
5371ccf
Refine DPO validation and docs
Copilot
authored
74e1b31
Polish DPO script formatting
Copilot
authored
1f3f463
Add opt-in profiling for continual DPO training
Copilot
authored
767b6a8
Refine continual DPO profiling instrumentation
Copilot
authored
bee90fe
Polish profiling typings and docs
Copilot
authored
91cdcf9
Harden profiling callback state handling
Copilot
authored
4acb486
Document profiling helper APIs
Copilot
authored
89a635c
Clarify continual DPO profiling error paths
Copilot
authored
4533610
Add profiling callback and timing docs
Copilot
authored
bfefda3
Add memory profiling regression tests
Copilot
authored
b17782a