Update README.md
Browse files
README.md
CHANGED
|
@@ -35,7 +35,7 @@ Training code is available here https://github.com/mozilla/smart-tab-grouping, t
|
|
| 35 |
The model has a strict input in the following formats.
|
| 36 |
|
| 37 |
```
|
| 38 |
-
Topic from keywords: [up to 3 comma-separated-lower-case-keywords]. titles \n [up to 3 \n separated titles]
|
| 39 |
```
|
| 40 |
|
| 41 |
Keywords is optional and should not be included for single tab use cases.
|
|
@@ -48,7 +48,7 @@ Topic from keywords: . titles \n Dogs - Google Search
|
|
| 48 |
```
|
| 49 |
|
| 50 |
```
|
| 51 |
-
Topic from keywords: dogs,food,pets. titles \
|
| 52 |
```
|
| 53 |
|
| 54 |
|
|
@@ -63,12 +63,14 @@ It filters some swear words. In some instances it may output 'None' when uncerta
|
|
| 63 |
|
| 64 |
## Training Details
|
| 65 |
|
| 66 |
-
Training data was created using OpenAI to create archetypes of 50 fake users and
|
| 67 |
|
| 68 |
Page titles from those synthetic pages were clustered, and then labeld using OpenAI, using an n-shot approach with hand-labeled examples in each query.
|
| 69 |
|
| 70 |
Training data was augmented with page titles extracted from thosands of English page titles Common Crawl dataset.
|
| 71 |
|
|
|
|
|
|
|
| 72 |
The training data was used to fine-tune a flan-t5-base model, which was later distilled on a t5-efficient-tiny model.
|
| 73 |
|
| 74 |
The model was then quantized to q8 (8 bit precision) for use in production Firefox.
|
|
|
|
| 35 |
The model has a strict input in the following formats.
|
| 36 |
|
| 37 |
```
|
| 38 |
+
Topic from keywords: [up to 3 comma-separated-lower-case-keywords]. titles: \n [up to 3 \n separated titles]
|
| 39 |
```
|
| 40 |
|
| 41 |
Keywords is optional and should not be included for single tab use cases.
|
|
|
|
| 48 |
```
|
| 49 |
|
| 50 |
```
|
| 51 |
+
Topic from keywords: dogs,food,pets. titles: \nDogs - Google Search\nDog Food - Shopping\nHow can I buy a pet - Google Search
|
| 52 |
```
|
| 53 |
|
| 54 |
|
|
|
|
| 63 |
|
| 64 |
## Training Details
|
| 65 |
|
| 66 |
+
Training data was created using OpenAI to create archetypes of 50 fake users and their imagined browsing activity for various tasks.
|
| 67 |
|
| 68 |
Page titles from those synthetic pages were clustered, and then labeld using OpenAI, using an n-shot approach with hand-labeled examples in each query.
|
| 69 |
|
| 70 |
Training data was augmented with page titles extracted from thosands of English page titles Common Crawl dataset.
|
| 71 |
|
| 72 |
+
An additional pre-processing step applied during training removes less important words from the training topic labels in order to shorten the topic to 1 word in most cases.
|
| 73 |
+
|
| 74 |
The training data was used to fine-tune a flan-t5-base model, which was later distilled on a t5-efficient-tiny model.
|
| 75 |
|
| 76 |
The model was then quantized to q8 (8 bit precision) for use in production Firefox.
|