Update README.md
Browse files
README.md
CHANGED
|
@@ -8,5 +8,39 @@ pipeline_tag: conversational
|
|
| 8 |
|
| 9 |
An auto-regressive causal LM created by combining 2x finetuned [Llama-2 70B](https://huggingface.co/meta-llama/llama-2-70b-hf) into one.
|
| 10 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 11 |
# Benchmarks
|
| 12 |
Coming soon.
|
|
|
|
| 8 |
|
| 9 |
An auto-regressive causal LM created by combining 2x finetuned [Llama-2 70B](https://huggingface.co/meta-llama/llama-2-70b-hf) into one.
|
| 10 |
|
| 11 |
+
# Prompting Format
|
| 12 |
+
|
| 13 |
+
Both Vicuna and Alpaca will work, but due the initial and final layers belonging primarily to Xwin, I expect Vicuna to work the best.
|
| 14 |
+
|
| 15 |
+
# Merge process
|
| 16 |
+
|
| 17 |
+
The models used in the merge are [Xwin](https://huggingface.co/Xwin-LM/Xwin-LM-70B-V0.1) and [Euryale](https://huggingface.co/Sao10K/Euryale-1.3-L2-70B).
|
| 18 |
+
|
| 19 |
+
The layer ranges used are as follows:
|
| 20 |
+
|
| 21 |
+
```yaml
|
| 22 |
+
- range 0, 16
|
| 23 |
+
Xwin
|
| 24 |
+
- range 8, 24
|
| 25 |
+
Euryale
|
| 26 |
+
- range 17, 32
|
| 27 |
+
Xwin
|
| 28 |
+
- range 25, 40
|
| 29 |
+
Euryale
|
| 30 |
+
- range 33, 48
|
| 31 |
+
Xwin
|
| 32 |
+
- range 41, 56
|
| 33 |
+
Euryale
|
| 34 |
+
- range 49, 64
|
| 35 |
+
Xwin
|
| 36 |
+
- range 57, 72
|
| 37 |
+
Euryale
|
| 38 |
+
- range 65, 80
|
| 39 |
+
Xwin
|
| 40 |
+
```
|
| 41 |
+
|
| 42 |
+
# Further details
|
| 43 |
+
Coming soon.
|
| 44 |
+
|
| 45 |
# Benchmarks
|
| 46 |
Coming soon.
|