File size: 3,335 Bytes
1d04e5b 7b1cbc7 5aac72a 3980f97 5aac72a 57b731d 9709984 94133a7 9709984 7b1cbc7 91ab63b ea9c357 91ab63b 62d891d 57b731d 91ab63b 1d04e5b |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 |
---
base_model:
- Sao10K/L3-8B-Stheno-v3.2
- NeverSleep/Llama-3-Lumimaid-8B-v0.1-OAS
library_name: transformers
tags:
- mergekit
- merge
---
<img src="https://huggingface.co/Alsebay/L3-8B-SMaid-v0.1/resolve/main/cover/cover.png" alt="img" style="width: 60%; min-width: 120px; height:80%; min-height: 200px; max-width:360px; max-height:600px; display: block">
> [!IMPORTANT]
> This model is wrongly using smaug-bpe pretokenizer during the merge (Mergekit uses smaug-bpe by default for llama-3 merges). So the quality of the model is worse than expected. Still usable with transformers loader.
# Update: Both version have different presents (settings) to work well
Overall:
Sao10K Stheno > SMaid V0.3 > SMaid V0.1 in Chai Benchmark
SMaid V0.1 = Sao10K Stheno > SMaid V0.3 in my custom EQ bench (Sadness and deep thought and Depression test)
Disclaimed: same seed, same character card, same scenario. 4 times try for each models.
# Best of L3-8B merge series for me. I choose 2 best variants to publish.
SMaid-V0.1: More smart, understand well content, more novelwriting. I like this version.
[SMaid-V0.3](https://huggingface.co/Alsebay/L3-8B-SMaid-v0.3): Upgrade from v0.1. More talkative, active, energetic (wrong setting, lol).
No V0.2 because I deleted it, it's a worst model of series.
I think Stheno and Lumumaid can be like a 'ying-yang', so I combine them, lol. Have test on Chaiverse, both of them got > 1995 elo score from begining. (Thanks Sao10K let me know about ChaiVerse :) )
SMaid = Stheno (it's very good) + LumiMaid (not too good, but the writing style is good)
**Recommend present (You can feedback if any setting is better)**
```
Temperature - 1.1-1.25
Min-P - 0.075
Top-K - 50
Top_P - 0.5
Repetition Penalty - 1.1
```
---
# Below is the auto-generate by Mergekit
This is a merge of pre-trained language models created using [mergekit](https://github.com/cg123/mergekit).
## Merge Details
### Merge Method
This model was merged using the [DARE](https://arxiv.org/abs/2311.03099) [TIES](https://arxiv.org/abs/2306.01708) merge method using [NeverSleep/Llama-3-Lumimaid-8B-v0.1-OAS](https://huggingface.co/NeverSleep/Llama-3-Lumimaid-8B-v0.1-OAS) as a base.
### Models Merged
The following models were included in the merge:
* [Sao10K/L3-8B-Stheno-v3.2](https://huggingface.co/Sao10K/L3-8B-Stheno-v3.2)
### Configuration
The following YAML configuration was used to produce this model:
```yaml
slices:
- sources:
- layer_range: [0, 16]
model: NeverSleep/Llama-3-Lumimaid-8B-v0.1-OAS
parameters:
density: 0.5
weight: 1.0
- layer_range: [0, 16]
model: Sao10K/L3-8B-Stheno-v3.2
parameters:
density: 0.5
weight: 0.9
- sources:
- layer_range: [16, 24]
model: Sao10K/L3-8B-Stheno-v3.2
parameters:
density: 0.75
weight: 0.5
- layer_range: [16, 24]
model: NeverSleep/Llama-3-Lumimaid-8B-v0.1-OAS
parameters:
density: 0.25
weight: 0.5
- sources:
- layer_range: [24, 32]
model: NeverSleep/Llama-3-Lumimaid-8B-v0.1-OAS
parameters:
density: 0.5
weight: 0.5
- layer_range: [24, 32]
model: Sao10K/L3-8B-Stheno-v3.2
parameters:
density: 0.5
weight: 1.0
merge_method: dare_ties
base_model: NeverSleep/Llama-3-Lumimaid-8B-v0.1-OAS
parameters:
int8_mask: true
dtype: bfloat16
```
|