Commit
•
dfd276b
1
Parent(s):
a5dcca3
Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,72 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
tags:
|
| 4 |
+
- merge
|
| 5 |
+
- mergekit
|
| 6 |
+
- lazymergekit
|
| 7 |
+
- AlekseiPravdin/KSI-RP-NSK-128k-7B
|
| 8 |
+
- grimjim/fireblossom-32K-7B
|
| 9 |
+
- gguf
|
| 10 |
+
- Q2_K
|
| 11 |
+
- Q3_K_L
|
| 12 |
+
- Q3_K_M
|
| 13 |
+
- Q3_K_S
|
| 14 |
+
- Q4_0
|
| 15 |
+
- Q4_1
|
| 16 |
+
- Q4_K_S
|
| 17 |
+
- Q4_k_m
|
| 18 |
+
- Q5_0
|
| 19 |
+
- Q5_1
|
| 20 |
+
- Q6_K
|
| 21 |
+
- Q5_K_S
|
| 22 |
+
- Q5_k_m
|
| 23 |
+
- Q8_0
|
| 24 |
+
- 128k
|
| 25 |
+
language:
|
| 26 |
+
- en
|
| 27 |
+
- ru
|
| 28 |
+
- th
|
| 29 |
+
---
|
| 30 |
+
|
| 31 |
+
# KSIF-RPG-128k-slerp ⭐️⭐️
|
| 32 |
+
|
| 33 |
+
KSIF-RPG-128k-slerp is a merge of the following models using [mergekit](https://github.com/cg123/mergekit):
|
| 34 |
+
* [AlekseiPravdin/KSI-RP-NSK-128k-7B](https://huggingface.co/AlekseiPravdin/KSI-RP-NSK-128k-7B)
|
| 35 |
+
* [grimjim/fireblossom-32K-7B](https://huggingface.co/grimjim/fireblossom-32K-7B)
|
| 36 |
+
|
| 37 |
+
## 🧩 Configuration
|
| 38 |
+
|
| 39 |
+
```yaml
|
| 40 |
+
slices:
|
| 41 |
+
- sources:
|
| 42 |
+
- model: AlekseiPravdin/KSI-RP-NSK-128k-7B
|
| 43 |
+
layer_range: [0, 32]
|
| 44 |
+
- model: grimjim/fireblossom-32K-7B
|
| 45 |
+
layer_range: [0, 32]
|
| 46 |
+
merge_method: slerp
|
| 47 |
+
base_model: AlekseiPravdin/KSI-RP-NSK-128k-7B
|
| 48 |
+
parameters:
|
| 49 |
+
t:
|
| 50 |
+
- filter: self_attn
|
| 51 |
+
value: [0, 0.5, 0.3, 0.7, 1]
|
| 52 |
+
- filter: mlp
|
| 53 |
+
value: [1, 0.5, 0.7, 0.3, 0]
|
| 54 |
+
- value: 0.5
|
| 55 |
+
dtype: bfloat16
|
| 56 |
+
|
| 57 |
+
```
|
| 58 |
+
|
| 59 |
+
Eval embedding benchmark (with 70 specific quesions):
|
| 60 |
+
|
| 61 |
+
|
| 62 |
+
|
| 63 |
+
|
| 64 |
+
|
| 65 |
+
|
| 66 |
+
|
| 67 |
+
|
| 68 |
+
|
| 69 |
+
|
| 70 |
+
|
| 71 |
+
|
| 72 |
+
|