Commit
•
3a8226c
1
Parent(s):
db8b03a
Delete readme.md
Browse files
readme.md
DELETED
|
@@ -1,61 +0,0 @@
|
|
| 1 |
-
|
| 2 |
-
This is my first "serious"(with practical use cases) experimental merge. Judge harshly. Mainly made for RP, but should be okay as an assistant. Turned out quite good, considering the amount of LORAs I merged into it.
|
| 3 |
-
|
| 4 |
-
# Observations
|
| 5 |
-
- GPTisms and repetition: put temperature and rep. pen. higher, make GPTisms stop sequences
|
| 6 |
-
- A bit different than the ususal stuff
|
| 7 |
-
- Lightly censored
|
| 8 |
-
- Fairly neutral, can be violent if you ask it really good, Goliath is a bit better at it
|
| 9 |
-
- Has a bit of optimism baked in, but it's not very severe
|
| 10 |
-
- Doesn't know when to stop, can be quite verbose or just stop almost immediately(maybe wants to use LimaRP settings idk)
|
| 11 |
-
- Sometimes can't handle '
|
| 12 |
-
- Second model that tried to be funny unprompted to me(First one was Goliath)
|
| 13 |
-
- Moderately intelligent
|
| 14 |
-
- Quite creative
|
| 15 |
-
|
| 16 |
-
# Naming
|
| 17 |
-
Internal name of this model was euryale-guano-saiga-med-janboros-kim-lima-wiz-tony-d30-s40, but I decided to keep it short, and since it was iteration G, I called it "Gembo".
|
| 18 |
-
|
| 19 |
-
# Prompt format
|
| 20 |
-
Alpaca. You can also try some other formats, I'm pretty sure it has a lot of them from all those merges.
|
| 21 |
-
```
|
| 22 |
-
### Instruction:
|
| 23 |
-
{instruction}
|
| 24 |
-
|
| 25 |
-
### Response:
|
| 26 |
-
```
|
| 27 |
-
|
| 28 |
-
# Settings
|
| 29 |
-
As I already mentioned, high temperature and rep.pen. works great.
|
| 30 |
-
For RP try something like this:
|
| 31 |
-
- temperature=5
|
| 32 |
-
- MinP=0.10
|
| 33 |
-
- rep.pen.=1.15
|
| 34 |
-
|
| 35 |
-
Adjust to match your needs.
|
| 36 |
-
|
| 37 |
-
|
| 38 |
-
# How it was created
|
| 39 |
-
I took Sao10K/Euryale-1.3-L2-70B (Good base model) and added
|
| 40 |
-
- Mikael110/llama-2-70b-guanaco-qlora (Creativity+assistant)
|
| 41 |
-
- IlyaGusev/saiga2_70b_lora (Creativity+assistant)
|
| 42 |
-
- s1ghhh/medllama-2-70b-qlora-1.1 (More data)
|
| 43 |
-
- v2ray/Airoboros-2.1-Jannie-70B-QLoRA (Creativity+assistant)
|
| 44 |
-
- Chat-Error/fiction.live-Kimiko-V2-70B (Creativity)
|
| 45 |
-
- Doctor-Shotgun/limarpv3-llama2-70b-qlora (Creativity)
|
| 46 |
-
- v2ray/LLaMA-2-Wizard-70B-QLoRA (Creativity+assistant)
|
| 47 |
-
- v2ray/TonyGPT-70B-QLoRA (Special spice)
|
| 48 |
-
|
| 49 |
-
Then I SLERP-merged it with cognitivecomputations/dolphin-2.2-70b (Needed to bridge the gap between this wonderful mess and Smaxxxer, otherwise it's quality is low) with 0.3t and then SLERP-merged it again with ChuckMcSneed/SMaxxxer-v1-70b (Creativity) with 0.4t. For SLERP-merges I used https://github.com/arcee-ai/mergekit.
|
| 50 |
-
|
| 51 |
-
# Benchmarks (Do they even mean anything anymore?)
|
| 52 |
-
### NeoEvalPlusN_benchmark
|
| 53 |
-
[My meme benchmark.](https://huggingface.co/datasets/ChuckMcSneed/NeoEvalPlusN_benchmark)
|
| 54 |
-
| Test name | Gembo |
|
| 55 |
-
| ---------- | ---------- |
|
| 56 |
-
| B | 2.5 |
|
| 57 |
-
| C | 1.5 |
|
| 58 |
-
| D | 3 |
|
| 59 |
-
| S | 7.5 |
|
| 60 |
-
| P | 5.25 |
|
| 61 |
-
| Total | 19.75 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|