Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,35 @@
|
|
| 1 |
---
|
| 2 |
license: apache-2.0
|
| 3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
license: apache-2.0
|
| 3 |
---
|
| 4 |
+
This is a linear model merge of:
|
| 5 |
+
|
| 6 |
+
60%
|
| 7 |
+
https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.2
|
| 8 |
+
|
| 9 |
+
40%
|
| 10 |
+
https://huggingface.co/stabilityai/japanese-stablelm-instruct-gamma-7b
|
| 11 |
+
|
| 12 |
+
|
| 13 |
+
## Evaluation
|
| 14 |
+
|
| 15 |
+
Tested on correct en-jp translation identification on the first 10k rows of https://huggingface.co/datasets/Verah/tatoeba_dedupe_en-jp_2024-March-01
|
| 16 |
+
|
| 17 |
+
Desired behaviour is to not accept any translation when we deliberaly test incorrect pairings from the dataset, and to not reject any translation when shown only correctly paired examples.
|
| 18 |
+
|
| 19 |
+
|
| 20 |
+
|Model | False Admissions| False Rejections |
|
| 21 |
+
|----------------|------------------|--------------------|
|
| 22 |
+
|Mistral Instruct | 41 | 600 |
|
| 23 |
+
|(This Model) | 13 | 1839 |
|
| 24 |
+
|JP Stable LM Gamma | 9679 | 138 |
|
| 25 |
+
|Hermes2DPO | 20 | 598 |
|
| 26 |
+
|
| 27 |
+
|
| 28 |
+
I made the test harder by concatenating 3 paired sentences together, in the false admissions case 1 out of those 3 was incorrectly paired.
|
| 29 |
+
|
| 30 |
+
|Model| False Admissions | False* Rejections|
|
| 31 |
+
|-----|------------------|-----------------|
|
| 32 |
+
|(This Model) | 89 | 5508|
|
| 33 |
+
|Hermes2DPO | 537| 1458|
|
| 34 |
+
|
| 35 |
+
This model also wanted to reject many "correct" translations, however 3 unrelated sentences back to back isn't a very correct thing to be doing, either.
|