Update README.md
Browse files
README.md
CHANGED
|
@@ -14,4 +14,78 @@ tags:
|
|
| 14 |
The BigWeave models aim to experimentally identify merge settings for increasing model performance. The version number merely tracks various attempts and is not a quality indicator. Only results demonstrating good performance are retained and shared.
|
| 15 |
|
| 16 |
# Prompting Format
|
| 17 |
-
Mistral, Vicuna and Alpaca.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 14 |
The BigWeave models aim to experimentally identify merge settings for increasing model performance. The version number merely tracks various attempts and is not a quality indicator. Only results demonstrating good performance are retained and shared.
|
| 15 |
|
| 16 |
# Prompting Format
|
| 17 |
+
Mistral, Vicuna and Alpaca.
|
| 18 |
+
|
| 19 |
+
Merge process
|
| 20 |
+
|
| 21 |
+
This is a self-merge of 152334H/miqu-1-70b-sf. By conducting exl2 measurements, we identify the most relevant layers. These layers are then duplicated in pairs to ensure overlaps.
|
| 22 |
+
|
| 23 |
+
Merge configuration:
|
| 24 |
+
```
|
| 25 |
+
slices:
|
| 26 |
+
- sources:
|
| 27 |
+
- model: 152334H/miqu-1-70b-sf
|
| 28 |
+
layer_range: [0,3]
|
| 29 |
+
- sources:
|
| 30 |
+
- model: 152334H/miqu-1-70b-sf
|
| 31 |
+
layer_range: [1,5]
|
| 32 |
+
- sources:
|
| 33 |
+
- model: 152334H/miqu-1-70b-sf
|
| 34 |
+
layer_range: [3,7]
|
| 35 |
+
- sources:
|
| 36 |
+
- model: 152334H/miqu-1-70b-sf
|
| 37 |
+
layer_range: [5,9]
|
| 38 |
+
- sources:
|
| 39 |
+
- model: 152334H/miqu-1-70b-sf
|
| 40 |
+
layer_range: [7,18]
|
| 41 |
+
- sources:
|
| 42 |
+
- model: 152334H/miqu-1-70b-sf
|
| 43 |
+
layer_range: [16,21]
|
| 44 |
+
- sources:
|
| 45 |
+
- model: 152334H/miqu-1-70b-sf
|
| 46 |
+
layer_range: [19,27]
|
| 47 |
+
- sources:
|
| 48 |
+
- model: 152334H/miqu-1-70b-sf
|
| 49 |
+
layer_range: [25,30]
|
| 50 |
+
- sources:
|
| 51 |
+
- model: 152334H/miqu-1-70b-sf
|
| 52 |
+
layer_range: [28,32]
|
| 53 |
+
- sources:
|
| 54 |
+
- model: 152334H/miqu-1-70b-sf
|
| 55 |
+
layer_range: [30,34]
|
| 56 |
+
- sources:
|
| 57 |
+
- model: 152334H/miqu-1-70b-sf
|
| 58 |
+
layer_range: [32,36]
|
| 59 |
+
- sources:
|
| 60 |
+
- model: 152334H/miqu-1-70b-sf
|
| 61 |
+
layer_range: [34,38]
|
| 62 |
+
- sources:
|
| 63 |
+
- model: 152334H/miqu-1-70b-sf
|
| 64 |
+
layer_range: [36,40]
|
| 65 |
+
- sources:
|
| 66 |
+
- model: 152334H/miqu-1-70b-sf
|
| 67 |
+
layer_range: [38,42]
|
| 68 |
+
- sources:
|
| 69 |
+
- model: 152334H/miqu-1-70b-sf
|
| 70 |
+
layer_range: [40,44]
|
| 71 |
+
- sources:
|
| 72 |
+
- model: 152334H/miqu-1-70b-sf
|
| 73 |
+
layer_range: [42,46]
|
| 74 |
+
- sources:
|
| 75 |
+
- model: 152334H/miqu-1-70b-sf
|
| 76 |
+
layer_range: [44,48]
|
| 77 |
+
- sources:
|
| 78 |
+
- model: 152334H/miqu-1-70b-sf
|
| 79 |
+
layer_range: [46,51]
|
| 80 |
+
- sources:
|
| 81 |
+
- model: 152334H/miqu-1-70b-sf
|
| 82 |
+
layer_range: [49,77]
|
| 83 |
+
- sources:
|
| 84 |
+
- model: 152334H/miqu-1-70b-sf
|
| 85 |
+
layer_range: [75,79]
|
| 86 |
+
- sources:
|
| 87 |
+
- model: 152334H/miqu-1-70b-sf
|
| 88 |
+
layer_range: [77,80]
|
| 89 |
+
merge_method: passthrough
|
| 90 |
+
dtype: float16
|
| 91 |
+
```
|