File size: 3,371 Bytes
1d04e5b 7b1cbc7 06beba9 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]
> Thank @mradermacher so much for help me find out that LumiMaid use 'smaug-bpe' pre-tokenizer. So that mean all its quant is unable to use. That mean you can only use Transformer to load this model for now (maybe they will fix or add feature in future)
# 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
```
|