Commit ·
c2b369a
1
Parent(s): b33d256
Update README.md
Browse files
README.md
CHANGED
|
@@ -1,9 +1,88 @@
|
|
| 1 |
---
|
| 2 |
library_name: peft
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 3 |
---
|
| 4 |
-
## Training procedure
|
| 5 |
|
| 6 |
-
#
|
| 7 |
|
|
|
|
| 8 |
|
| 9 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
library_name: peft
|
| 3 |
+
license: mit
|
| 4 |
+
datasets:
|
| 5 |
+
- iva_mt_wslot
|
| 6 |
+
metrics:
|
| 7 |
+
- bleu
|
| 8 |
+
model-index:
|
| 9 |
+
- name: iva_mt_wslot-m2m100_418M-en-pl-lora_adapter
|
| 10 |
+
results:
|
| 11 |
+
- task:
|
| 12 |
+
name: Sequence-to-sequence Language Modeling
|
| 13 |
+
type: text2text-generation
|
| 14 |
+
dataset:
|
| 15 |
+
name: iva_mt_wslot
|
| 16 |
+
type: iva_mt_wslot
|
| 17 |
+
config: en-pl
|
| 18 |
+
split: validation
|
| 19 |
+
args: en-pl
|
| 20 |
+
metrics:
|
| 21 |
+
- name: Bleu
|
| 22 |
+
type: bleu
|
| 23 |
+
value: 38.2365
|
| 24 |
+
language:
|
| 25 |
+
- pl
|
| 26 |
+
tags:
|
| 27 |
+
- machine translation
|
| 28 |
+
- iva
|
| 29 |
+
- virtual assistants
|
| 30 |
+
- natural language understanding
|
| 31 |
+
- nlu
|
| 32 |
---
|
|
|
|
| 33 |
|
| 34 |
+
# (WIP!) iva_mt_wslot-m2m100_418M-en-pl-lora_adapter
|
| 35 |
|
| 36 |
+
Notice: **Although training results are good for some reason inference results are rather poor. I'm leaving this model here as a PoC that PERF LORA adaptation for M2M100 is possible.**
|
| 37 |
|
| 38 |
+
This model is a LORA adapted version of [facebook/m2m100_418M](https://huggingface.co/facebook/m2m100_418M) on the iva_mt_wslot dataset.
|
| 39 |
+
It achieves the following results on the test set (measured with sacrebleu):
|
| 40 |
+
- Bleu: 9.33
|
| 41 |
+
|
| 42 |
+
## Using
|
| 43 |
+
|
| 44 |
+
The model can be used as follows:
|
| 45 |
+
|
| 46 |
+
First, clone the repository and navigate to the project directory:
|
| 47 |
+
|
| 48 |
+
```bash
|
| 49 |
+
git clone https://github.com/cartesinus/multiverb_iva_mt
|
| 50 |
+
cd multiverb_iva_mt
|
| 51 |
+
```
|
| 52 |
+
|
| 53 |
+
Then:
|
| 54 |
+
|
| 55 |
+
```python
|
| 56 |
+
import csv
|
| 57 |
+
from iva_mt.iva_mt import IVAMT
|
| 58 |
+
import pandas as pd
|
| 59 |
+
|
| 60 |
+
lang = "es"
|
| 61 |
+
translator = IVAMT(lang, peft_model_id="cartesinus/iva_mt_wslot-m2m100_418M-en-es-lora_adapter", device="cuda:0", batch_size=128)
|
| 62 |
+
trans = translator.translate("here your example")[0]
|
| 63 |
+
```
|
| 64 |
+
|
| 65 |
+
## Training results
|
| 66 |
+
|
| 67 |
+
| Epoch | Training Loss | Validation Loss | Bleu | Gen Len |
|
| 68 |
+
|:-----:|:-------------:|:---------------:|:-------:|:-------:|
|
| 69 |
+
| 1 | 7.8621 | 7.6870 | 24.9063 | 19.3322 |
|
| 70 |
+
| 2 | 7.6340 | 7.5312 | 29.7956 | 19.7533 |
|
| 71 |
+
| 3 | 7.5582 | 7.4595 | 34.8184 | 20.1269 |
|
| 72 |
+
| 4 | 7.5047 | 7.4264 | 36.1874 | 20.5621 |
|
| 73 |
+
| 5 | 7.4888 | 7.4167 | 36.2287 | 20.4417 |
|
| 74 |
+
| 6 | 7.4560 | 7.4013 | 36.6355 | 20.2241 |
|
| 75 |
+
| 7 | 7.4477 | 7.3907 | 37.0554 | 20.0945 |
|
| 76 |
+
| 8 | 7.4422 | 7.3743 | 37.7549 | 20.1589 |
|
| 77 |
+
| 9 | 7.4311 | 7.3748 | 37.5705 | 19.9370 |
|
| 78 |
+
| 10 | 7.4294 | 7.3679 | 37.5343 | 20.2241 |
|
| 79 |
+
| 11 | 7.4114 | 7.3697 | 38.1872 | 20.3836 |
|
| 80 |
+
| 12 | 7.4224 | 7.3620 | 38.1759 | 20.1785 |
|
| 81 |
+
| 13 | 7.4334 | 7.3608 | 38.0895 | 20.2996 |
|
| 82 |
+
| 14 | 7.4133 | 7.3621 | 38.2365 | 20.2948 |
|
| 83 |
+
| 15 | 7.4158 | 7.3599 | 38.1056 | 20.2010 |
|
| 84 |
+
|
| 85 |
+
|
| 86 |
+
## Framework versions
|
| 87 |
+
|
| 88 |
+
- PEFT 0.5.0
|