Commit
•
5d7eee1
1
Parent(s):
b3605d0
Update README.md
Browse files
README.md
CHANGED
@@ -4,13 +4,14 @@ datasets:
|
|
4 |
- EleutherAI/wikitext_document_level
|
5 |
language:
|
6 |
- en
|
|
|
7 |
---
|
8 |
|
9 |
[<img src="https://raw.githubusercontent.com/OpenAccess-AI-Collective/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>](https://github.com/OpenAccess-AI-Collective/axolotl)
|
10 |
|
11 |
LLaMA 33b finetuned on `wikitext_document_level` with a combination of both linear and NTK-aware ROPE scaling.
|
12 |
|
13 |
-
Trained with alpha=4, scale=2.
|
14 |
|
15 |
<img src="llama33b-s2a4-qlora/resolve/main/perplexity.png" alt="Perplexity Graph" />
|
16 |
|
@@ -30,4 +31,4 @@ The following `bitsandbytes` quantization config was used during training:
|
|
30 |
### Framework versions
|
31 |
|
32 |
|
33 |
-
- PEFT 0.4.0.dev0
|
|
|
4 |
- EleutherAI/wikitext_document_level
|
5 |
language:
|
6 |
- en
|
7 |
+
pipeline_tag: text-generation
|
8 |
---
|
9 |
|
10 |
[<img src="https://raw.githubusercontent.com/OpenAccess-AI-Collective/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>](https://github.com/OpenAccess-AI-Collective/axolotl)
|
11 |
|
12 |
LLaMA 33b finetuned on `wikitext_document_level` with a combination of both linear and NTK-aware ROPE scaling.
|
13 |
|
14 |
+
Trained with alpha=4, scale=2. Definitely works for sequence lengths up to and including 4096. Might work for much longer, but I don't have the VRAM to test properly. ¯\\\_(ツ)\_/¯
|
15 |
|
16 |
<img src="llama33b-s2a4-qlora/resolve/main/perplexity.png" alt="Perplexity Graph" />
|
17 |
|
|
|
31 |
### Framework versions
|
32 |
|
33 |
|
34 |
+
- PEFT 0.4.0.dev0
|