Updated evals with newer ones and larger size
Browse files
README.md
CHANGED
@@ -19,32 +19,43 @@ The base OPT-2.7B model is licensed under the OPT-175B license, Copyright (c) Me
|
|
19 |
# Evaluation Results
|
20 |
As the original datasets used for the source models are not publically available, I use my own datasets for this evaluation, which may not provide accurate comparison.
|
21 |
|
22 |
-
Eval parameters:
|
23 |
|
24 |
```
|
25 |
Literotica Dataset Eval (Randomly selected stories)
|
26 |
-
{'eval_loss': 2.
|
27 |
-
{'eval_loss': 2.
|
28 |
-
{'eval_loss': 2.
|
29 |
-
{'eval_loss': 2.
|
|
|
30 |
|
31 |
ASSTR Dataset Eval (Randomly selected stories)
|
32 |
-
{'eval_loss': 2.
|
33 |
-
{'eval_loss': 2.
|
34 |
-
{'eval_loss': 2.
|
35 |
-
{'eval_loss': 2.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
36 |
|
37 |
Harry Potter Dataset Eval (Canon books)
|
38 |
-
{'eval_loss': 2.
|
39 |
-
{'eval_loss': 2.
|
40 |
-
{'eval_loss': 2.
|
41 |
-
{'eval_loss': 2.
|
|
|
42 |
|
43 |
Star Wars Dataset Eval (Rogue One Novel)
|
44 |
-
{'eval_loss': 2.
|
45 |
-
{'eval_loss': 2.
|
46 |
-
{'eval_loss': 2.
|
47 |
-
{'eval_loss': 2.
|
|
|
48 |
|
49 |
```
|
50 |
|
19 |
# Evaluation Results
|
20 |
As the original datasets used for the source models are not publically available, I use my own datasets for this evaluation, which may not provide accurate comparison.
|
21 |
|
22 |
+
Eval parameters: 32000 characters extracted from the middle of the corpus, tested in blocks of 1024 tokens each, same dataset used for each test batch.
|
23 |
|
24 |
```
|
25 |
Literotica Dataset Eval (Randomly selected stories)
|
26 |
+
{'eval_loss': 2.571258306503296, 'name': 'Concedo_OPT-2.7B-Nerybus-Mix'}
|
27 |
+
{'eval_loss': 2.5491442680358887, 'name': 'KoboldAI_OPT-2.7B-Erebus'}
|
28 |
+
{'eval_loss': 2.6158597469329834, 'name': 'KoboldAI_OPT-2.7B-Nerys'}
|
29 |
+
{'eval_loss': 2.614469051361084, 'name': 'facebook_opt-2.7b'}
|
30 |
+
{'eval_loss': 2.4960227012634277, 'name': '(Unreleased 2.7B ModronAI Model)'}
|
31 |
|
32 |
ASSTR Dataset Eval (Randomly selected stories)
|
33 |
+
{'eval_loss': 2.664412498474121, 'name': 'Concedo_OPT-2.7B-Nerybus-Mix'}
|
34 |
+
{'eval_loss': 2.6451029777526855, 'name': 'KoboldAI_OPT-2.7B-Erebus'}
|
35 |
+
{'eval_loss': 2.7259647846221924, 'name': 'KoboldAI_OPT-2.7B-Nerys'}
|
36 |
+
{'eval_loss': 2.6675195693969727, 'name': 'facebook_opt-2.7b'}
|
37 |
+
{'eval_loss': 2.962111473083496, 'name': '(Unreleased 2.7B ModronAI Model)'}
|
38 |
+
|
39 |
+
Sexstories Dataset Eval (Random highly rated stories)
|
40 |
+
{'eval_loss': 2.2352423667907715, 'name': 'Concedo_OPT-2.7B-Nerybus-Mix'}
|
41 |
+
{'eval_loss': 2.194378137588501, 'name': 'KoboldAI_OPT-2.7B-Erebus'}
|
42 |
+
{'eval_loss': 2.307469129562378, 'name': 'KoboldAI_OPT-2.7B-Nerys'}
|
43 |
+
{'eval_loss': 2.293961763381958, 'name': 'facebook_opt-2.7b'}
|
44 |
+
{'eval_loss': 2.0103421211242676, 'name': '(Unreleased 2.7B ModronAI Model)'}
|
45 |
|
46 |
Harry Potter Dataset Eval (Canon books)
|
47 |
+
{'eval_loss': 2.473742961883545, 'name': 'Concedo_OPT-2.7B-Nerybus-Mix'}
|
48 |
+
{'eval_loss': 2.480600357055664, 'name': 'KoboldAI_OPT-2.7B-Erebus'}
|
49 |
+
{'eval_loss': 2.506237506866455, 'name': 'KoboldAI_OPT-2.7B-Nerys'}
|
50 |
+
{'eval_loss': 2.5074169635772705, 'name': 'facebook_opt-2.7b'}
|
51 |
+
{'eval_loss': 2.273703098297119, 'name': '(Unreleased 2.7B ModronAI Model)'}
|
52 |
|
53 |
Star Wars Dataset Eval (Rogue One Novel)
|
54 |
+
{'eval_loss': 2.5031676292419434, 'name': 'Concedo_OPT-2.7B-Nerybus-Mix'}
|
55 |
+
{'eval_loss': 2.5239150524139404, 'name': 'KoboldAI_OPT-2.7B-Erebus'}
|
56 |
+
{'eval_loss': 2.526801586151123, 'name': 'KoboldAI_OPT-2.7B-Nerys'}
|
57 |
+
{'eval_loss': 2.473283529281616, 'name': 'facebook_opt-2.7b'}
|
58 |
+
{'eval_loss': 2.955465793609619, 'name': '(Unreleased 2.7B ModronAI Model)'}
|
59 |
|
60 |
```
|
61 |
|