Update README.md
Browse files
README.md
CHANGED
|
@@ -1,5 +1,4 @@
|
|
| 1 |
---
|
| 2 |
-
|
| 3 |
language:
|
| 4 |
- en
|
| 5 |
tags:
|
|
@@ -8,8 +7,7 @@ tags:
|
|
| 8 |
- gpt
|
| 9 |
license: mit
|
| 10 |
datasets:
|
| 11 |
-
-
|
| 12 |
-
|
| 13 |
widget:
|
| 14 |
- text: "Do you like my new haircut?\nperson beta:\n\n"
|
| 15 |
example_title: "haircut"
|
|
@@ -19,7 +17,6 @@ widget:
|
|
| 19 |
example_title: "favorite"
|
| 20 |
- text: "how much does it cost?\nperson beta:\n\n"
|
| 21 |
example_title: "money"
|
| 22 |
-
|
| 23 |
inference:
|
| 24 |
parameters:
|
| 25 |
min_length: 2
|
|
@@ -30,12 +27,10 @@ inference:
|
|
| 30 |
top_p: 0.85
|
| 31 |
top_k: 10
|
| 32 |
repetition_penalty: 2.1
|
| 33 |
-
|
| 34 |
-
|
| 35 |
---
|
| 36 |
-
# ai-msgbot GPT2-XL-dialogue
|
| 37 |
|
| 38 |
-
|
|
|
|
| 39 |
|
| 40 |
GPT2-XL (~1.5 B parameters) trained on [the Wizard of Wikipedia dataset](https://parl.ai/projects/wizard_of_wikipedia/) for 40k steps with **33**/36 layers frozen using `aitextgen`. The resulting model was then **further fine-tuned** on the [Daily Dialogues](http://yanran.li/dailydialog) for 40k steps, with **34**/36 layers frozen.
|
| 41 |
|
|
|
|
| 1 |
---
|
|
|
|
| 2 |
language:
|
| 3 |
- en
|
| 4 |
tags:
|
|
|
|
| 7 |
- gpt
|
| 8 |
license: mit
|
| 9 |
datasets:
|
| 10 |
+
- natural_questions
|
|
|
|
| 11 |
widget:
|
| 12 |
- text: "Do you like my new haircut?\nperson beta:\n\n"
|
| 13 |
example_title: "haircut"
|
|
|
|
| 17 |
example_title: "favorite"
|
| 18 |
- text: "how much does it cost?\nperson beta:\n\n"
|
| 19 |
example_title: "money"
|
|
|
|
| 20 |
inference:
|
| 21 |
parameters:
|
| 22 |
min_length: 2
|
|
|
|
| 27 |
top_p: 0.85
|
| 28 |
top_k: 10
|
| 29 |
repetition_penalty: 2.1
|
|
|
|
|
|
|
| 30 |
---
|
|
|
|
| 31 |
|
| 32 |
+
# ai-msgbot: GPT2-XL-dialogue
|
| 33 |
+
|
| 34 |
|
| 35 |
GPT2-XL (~1.5 B parameters) trained on [the Wizard of Wikipedia dataset](https://parl.ai/projects/wizard_of_wikipedia/) for 40k steps with **33**/36 layers frozen using `aitextgen`. The resulting model was then **further fine-tuned** on the [Daily Dialogues](http://yanran.li/dailydialog) for 40k steps, with **34**/36 layers frozen.
|
| 36 |
|