53 lines
1.5 KiB
Markdown
53 lines
1.5 KiB
Markdown
|
---
|
||
|
|
license: cc-by-nc-4.0
|
||
|
|
tags:
|
||
|
|
- merge
|
||
|
|
- conversational
|
||
|
|
- multi-task
|
||
|
|
pipeline_tag: text-generation
|
||
|
|
---
|
||
|
|
|
||
|
|
# Winter Garden 7B - β - "The Writer"
|
||
|
|
|
||
|
|
It was mentioned that we are in the open ai dark winter; so I thought I would make myself a nice winter garden.
|
||
|
|
|
||
|
|
## An experiment
|
||
|
|
|
||
|
|
I've merged four partitions successfully in the past, so lets go for 9! I started with:
|
||
|
|
|
||
|
|
* Mistral-7B-v0.1
|
||
|
|
|
||
|
|
and merged in
|
||
|
|
|
||
|
|
* ZySec-7B-v1
|
||
|
|
* LemonadeRP-4.5.3
|
||
|
|
* dpo-binarized-NeutrixOmnibe-7B
|
||
|
|
* Multi-Verse-RP-7B
|
||
|
|
* AlphaMonarch-7B
|
||
|
|
* opus-v1.2-7b
|
||
|
|
* Kunoichi-DPO-v2-7B
|
||
|
|
* Noromaid-7B-0.4-DPO
|
||
|
|
* ogno-monarch-jaskier-merge-7b-OH-PREF-DPO-v2
|
||
|
|
|
||
|
|
### 9-partition merge
|
||
|
|
|
||
|
|
All of the layers were partitioned in to 9 random bins. Alternating models were slerped at [0...1], and [1...0] gradients; except attention, which was slerped at 0.03.
|
||
|
|
|
||
|
|
This means that the model is still predominantly ordered around base mistral - including half of the input and output layers, and 28% of attention.
|
||
|
|
|
||
|
|
### Other
|
||
|
|
|
||
|
|
Includes fast tokenizer.
|
||
|
|
|
||
|
|
## Chat Template
|
||
|
|
|
||
|
|
There is currently a chat template for conversational turns. It may or may not be good for this model without more tuning. Initial tests show this model is a real talker and writer.
|
||
|
|
|
||
|
|
* If you prompt with ```[WP] <prompt>\n\n``` it will take right off.
|
||
|
|
|
||
|
|
It doesn't seem to know `### Instruction:` or `<|im_start|>`; `<|user|>` ,`<|assistant|>` works but tends to get quickly mixed up with html, and back and forth conversation devolves in to long narratives.
|
||
|
|
|
||
|
|
## Scores
|
||
|
|
|
||
|
|
Metric | Score
|
||
|
|
---|---
|