Update README.md
Browse files
README.md
CHANGED
@@ -6,5 +6,21 @@ datasets:
|
|
6 |
- garage-bAInd/Open-Platypus
|
7 |
---
|
8 |
|
|
|
9 |
|
10 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
6 |
- garage-bAInd/Open-Platypus
|
7 |
---
|
8 |
|
9 |
+
# Instruction tune of Yi-34b with Open-Platypus (fp16)
|
10 |
|
11 |
+
|
12 |
+
## Overview
|
13 |
+
|
14 |
+
This is [chargoddard/Yi-34B-Llama](https://huggingface.co/chargoddard/Yi-34B-Llama), with instruction tuning performed with the [garage-bAInd/Open-Platypus](https://huggingface.co/datasets/garage-bAInd/Open-Platypus) dataset. That base model is [01-ai/Yi-34B](https://huggingface.co/01-ai/Yi-34B), but using llama2 model definitions and tokenizer to remove any remote code requirements.
|
15 |
+
|
16 |
+
**This is a (merged) QLoRA fine-tune (rank 64)**.
|
17 |
+
|
18 |
+
The finetune was performed with 1x RTX 6000 Ada (~18 hours to this checkpoint). It is possible this is rather undertrained, as this checkpoint is at 1 epoch. I began to see some performance degradation after that; more hyperparameter tuning is probably warranted.
|
19 |
+
|
20 |
+
## How to Use
|
21 |
+
|
22 |
+
Use as you would any llama-2 model.
|
23 |
+
|
24 |
+
## Prompting:
|
25 |
+
|
26 |
+
Model was trained with legacy airoboros <2.0 system prompt. See [bhenrym14/airoboros-33b-gpt4-1.4.1-lxctx-PI-16384-fp16](https://huggingface.co/bhenrym14/airoboros-33b-gpt4-1.4.1-lxctx-PI-16384-fp16) model card for details.
|