Commit
•
5ce3a7a
1
Parent(s):
e3563a6
Update README.md
Browse files
README.md
CHANGED
@@ -24,6 +24,10 @@ This Yi model is "Llama-fied", meaning the keys are renamed to match those used
|
|
24 |
|
25 |
This model is retrained thanks to compute provided by [alpin](https://huggingface.co/alpindale) with a monkeypatch to the trainer to resolve EOS token issues in the prompter. A smaller batch size and learning rate were used and training was extended by one epoch. 8-bit lora was also used instead of qlora.
|
26 |
|
|
|
|
|
|
|
|
|
27 |
## Usage:
|
28 |
|
29 |
The intended prompt format is the modified Vicuna 1.1 instruction format used by airoboros v2:
|
|
|
24 |
|
25 |
This model is retrained thanks to compute provided by [alpin](https://huggingface.co/alpindale) with a monkeypatch to the trainer to resolve EOS token issues in the prompter. A smaller batch size and learning rate were used and training was extended by one epoch. 8-bit lora was also used instead of qlora.
|
26 |
|
27 |
+
## EXL2 Quants Provided by LoneStriker:
|
28 |
+
|
29 |
+
[3.0bpw](https://huggingface.co/LoneStriker/airoboros-2.2.1-y34b-3.0bpw-h6-exl2), [4.0bpw](https://huggingface.co/LoneStriker/airoboros-2.2.1-y34b-4.0bpw-h6-exl2), [5.0bpw](https://huggingface.co/LoneStriker/airoboros-2.2.1-y34b-5.0bpw-h6-exl2), [6.0bpw](https://huggingface.co/LoneStriker/airoboros-2.2.1-y34b-6.0bpw-h6-exl2), [8.0bpw](https://huggingface.co/LoneStriker/airoboros-2.2.1-y34b-8.0bpw-h8-exl2)
|
30 |
+
|
31 |
## Usage:
|
32 |
|
33 |
The intended prompt format is the modified Vicuna 1.1 instruction format used by airoboros v2:
|