Downloads · 30 days
0
w2vbert2-lin-waxal
Abdoul27/w2vbert2-lin-waxal
w2vbert2-lin-waxal is a machine learning model from Abdoul27. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
facebook/w2v-bert-2.0 with a fresh 43-symbol character head, 2500 steps (4.9 epochs) over 16,320 WAXAL Lingala clips, 85.2 h. Training text is lowercased with punctuation stripped.
Access
Public
Updated Aug 3, 2026
Repo size
4.8 GB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors4.8 GB · 100%
At a glance
- Access
- Public
- Created
- Aug 3, 2026
- Updated
- Aug 3, 2026
- SHA
- fe590394
Model card
From the Hugging Face model README
w2v-BERT 2.0 -> WAXAL Lingala CTC
facebook/w2v-bert-2.0 with a fresh 43-symbol character head, 2500 steps
(4.9 epochs) over 16,320 WAXAL Lingala clips, 85.2 h. Training text is
lowercased with punctuation stripped.
| checkpoint | decode | Zindi Lingala-half probe |
|---|---|---|
| 2000 | greedy | 0.280922 |
| 2000 | + KenLM 5-gram (alpha 0.060, beta -0.031, beam 200) | 0.287731 |
LM: WAXAL + AfriVoice Lingala, 35,120 sentences, unigrams filtered to the model alphabet. 5-gram beat 4-gram (0.8108) and 3-gram (0.8105) on held-out tuning data.
checkpoint-2000 is best by agreement screening (0.847 vs the reference
system, against 0.823 for 2500).
- Created
- Aug 3, 2026
- Updated
- Aug 3, 2026