Spaces:
Running
Running
Upload folder using huggingface_hub
Browse files- sessions/2026-05-18.md +48 -0
sessions/2026-05-18.md
ADDED
|
@@ -0,0 +1,48 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# Session Notes — 18 May 2026
|
| 2 |
+
|
| 3 |
+
## Summary
|
| 4 |
+
Reduced project from 7 to 4 languages (Swahili, Hausa, Yoruba, Nyanja) per supervisor directive. Pulled live annotation data from HF Space. Confirmed Hausa Nigeria (Kano) annotations complete.
|
| 5 |
+
|
| 6 |
+
## What We Did
|
| 7 |
+
|
| 8 |
+
### 1. Pulled Live Annotation Data from HF Space
|
| 9 |
+
- Added `/admin/export-db` endpoint to download the live database
|
| 10 |
+
- Added `/admin/users/{username}` DELETE endpoint for user management
|
| 11 |
+
- Downloaded `afridialeval_live.db` (856KB) for local analysis
|
| 12 |
+
|
| 13 |
+
### 2. Reviewed Annotation Progress
|
| 14 |
+
| Region | Status |
|
| 15 |
+
|--------|--------|
|
| 16 |
+
| Hausa Nigeria (Kano) | ✅ DONE — 3/3 annotators × 5/5 dialogues |
|
| 17 |
+
| Swahili Kenya (Nairobi) | 🔄 In progress — Isaac done, Grace 1/3, Timothy 0/5 |
|
| 18 |
+
|
| 19 |
+
**Hausa Kano Results:**
|
| 20 |
+
- nyati (gemini-3-flash-preview): avg naturalness 4.2, preferred 14/15
|
| 21 |
+
- kifaru (gpt-5.1): avg naturalness 3.6, preferred 1/15
|
| 22 |
+
- chui (gemma-3-27b-it): avg naturalness 2.4
|
| 23 |
+
- tembo (qwen-3.5-27b): avg naturalness 1.13
|
| 24 |
+
|
| 25 |
+
### 3. Removed Languages — Reduced to 4
|
| 26 |
+
- **Removed:** Somali, Tsonga, Amharic (all regions)
|
| 27 |
+
- **Kept:** Swahili, Hausa, Yoruba, Nyanja
|
| 28 |
+
- Moved 72 files to `data/removed/` (local only, gitignored)
|
| 29 |
+
- Removed from `seed_users.sh`, `src/region_registry.py`
|
| 30 |
+
- Purged variants/tasks/annotators from live HF database
|
| 31 |
+
- Updated `import_data.py` to auto-purge inactive languages on startup
|
| 32 |
+
|
| 33 |
+
### 4. Removed Bessy Mukaria
|
| 34 |
+
- Deleted from live HF DB (withdrew from project)
|
| 35 |
+
|
| 36 |
+
## Still Pending
|
| 37 |
+
- **Yoruba annotators needed** — both Nigeria-Lagos and Benin-Porto-Novo regions
|
| 38 |
+
- **Swahili Kenya** — follow up with Timothy Ngao (0/5) and Grace Ngari (1/3)
|
| 39 |
+
- Remaining regions: Swahili Tanzania-Zanzibar, Nyanja Mozambique-Tete
|
| 40 |
+
|
| 41 |
+
## Commits Today
|
| 42 |
+
| Hash | Description |
|
| 43 |
+
|------|-------------|
|
| 44 |
+
| `327e52a` | Add admin DB export endpoint for data retrieval |
|
| 45 |
+
| `54ff1d2` | Add admin user deletion endpoint |
|
| 46 |
+
| `c8697c7` | Remove Somali, Tsonga, Amharic — reduce to 4 languages |
|
| 47 |
+
| `42e1506` | Add data/removed/ to .gitignore and untrack from repo |
|
| 48 |
+
| `c4e671f` | Purge removed languages from DB on import |
|