feat: add record-mode to Store for DPO/GRPO

- Store gains fetch_record/num_records alongside stream fetch/__len__
- save_bin/load_bin support per-record offsets via record_keys param
- H5Store/MmapStore/JsonlStore all support dual stream+record access
- DPODataset/GRPODataset use fetch_record, no cross-record concat
- dpo_collate_fn + collate_fn wired through TrainConfig
- fixes attention context leakage in DPO from windowed concatenation
This commit is contained in:
2026-07-18 21:02:29 +08:00
parent 28886e4241
commit a74e5b91a3
12 changed files with 334 additions and 92 deletions
+2 -2
View File
@@ -12,7 +12,7 @@ from astrai.config import (
from astrai.dataset import (
BaseDataset,
DatasetFactory,
ResumableDistributedSampler,
RDSampler,
Store,
StoreFactory,
)
@@ -77,7 +77,7 @@ __all__ = [
"Pipeline",
"PipelineConfig",
"ProtocolHandler",
"ResumableDistributedSampler",
"RDSampler",
"SamplingPipeline",
"SchedulerFactory",
"Store",