Fix README with correct datasets/splits and curriculum notes
Browse files
README.md
CHANGED
|
@@ -14,18 +14,18 @@ datasets:
|
|
| 14 |
- zero9tech/data-scientist-insight-dialog-en-16.5k
|
| 15 |
---
|
| 16 |
|
| 17 |
-
# QWEN3-4B-Data-Science-V2
|
| 18 |
|
| 19 |
## Training Setup
|
| 20 |
-
Domain SFT on .
|
| 21 |
|
| 22 |
## Dataset Snapshot
|
| 23 |
-
- Total records:
|
| 24 |
-
- Split:
|
| 25 |
-
- Hard gate:
|
| 26 |
-
- Soft gate:
|
| 27 |
-
- Self-BLEU (final):
|
| 28 |
-
- Distinct-2 (final):
|
| 29 |
|
| 30 |
## Copyright
|
| 31 |
Copyright (c) Zero9 Tech
|
|
|
|
| 14 |
- zero9tech/data-scientist-insight-dialog-en-16.5k
|
| 15 |
---
|
| 16 |
|
| 17 |
+
# QWEN3-4B-Data-Science-V2 - Insight Navigator
|
| 18 |
|
| 19 |
## Training Setup
|
| 20 |
+
Domain SFT on zero9tech/data-scientist-insight-dialog-en-16.5k.
|
| 21 |
|
| 22 |
## Dataset Snapshot
|
| 23 |
+
- Total records: 16,463
|
| 24 |
+
- Split: train 14,021 ; validation 801 ; test 1,641
|
| 25 |
+
- Hard gate: PASS
|
| 26 |
+
- Soft gate: PASS
|
| 27 |
+
- Self-BLEU (final): 0.5574
|
| 28 |
+
- Distinct-2 (final): 0.0768
|
| 29 |
|
| 30 |
## Copyright
|
| 31 |
Copyright (c) Zero9 Tech
|