Skip to content

Commit e91c743

Browse files
jioffe502claude
andcommitted
docs: update HANDOFF.md with store harness integration
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Signed-off-by: Jacob Ioffe <jioffe@nvidia.com>
1 parent 6d018dc commit e91c743

File tree

1 file changed

+13
-0
lines changed

1 file changed

+13
-0
lines changed

nemo_retriever/harness/HANDOFF.md

Lines changed: 13 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -153,6 +153,16 @@ Notes:
153153
- Comparison utility:
154154
- `retriever harness compare` prints pages/sec and recall deltas by run name for two sessions.
155155

156+
## Recent Progress (Apr 2026)
157+
158+
- Image and text storage is now wired end-to-end through the harness:
159+
- Harness config fields: `store_images_uri`, `store_text`, `strip_base64`.
160+
- Environment overrides: `HARNESS_STORE_IMAGES_URI`, `HARNESS_STORE_TEXT`, `HARNESS_STRIP_BASE64`.
161+
- `_build_command()` maps these to batch CLI flags (`--store-images-uri`, `--store-text`, `--strip-base64` / `--no-strip-base64`).
162+
- When `store_images_uri` is set, harness resolves it relative to the artifact directory so stored outputs land alongside other run artifacts.
163+
- Defaults: `store_images_uri: null`, `store_text: false`, `strip_base64: true`.
164+
- Store metadata (`store_images_uri`, `store_text`, `strip_base64`) is recorded in `results.json`.
165+
156166
## Current Validation Status
157167

158168
Harness-focused tests pass:
@@ -187,6 +197,9 @@ shim for that newer upstream behavior.
187197
- `gpu_page_elements` -> `--page-elements-gpus-per-actor`
188198
- `gpu_ocr` -> `--ocr-gpus-per-actor`
189199
- `gpu_embed` -> `--embed-gpus-per-actor`
200+
- `store_images_uri` -> `--store-images-uri`
201+
- `store_text` -> `--store-text`
202+
- `strip_base64` -> `--strip-base64` / `--no-strip-base64`
190203

191204
### Artifact / parser semantics
192205

0 commit comments

Comments
 (0)