d37efa31cf
Model: Indexnusrefather/Super-Slop-Machina-Roleplay-1.2b-ERP-Tolerant-V2 Source: Original Platform
57 lines
1.9 KiB
Markdown
57 lines
1.9 KiB
Markdown
---
|
|
license: other
|
|
license_name: lfm1.0
|
|
license_link: LICENSE
|
|
language:
|
|
- en
|
|
base_model:
|
|
- LiquidAI/LFM2.5-1.2B-Instruct
|
|
pipeline_tag: text-generation
|
|
tags:
|
|
- rp
|
|
- roleplay
|
|
- creative
|
|
- writer
|
|
- finetune
|
|
- lora
|
|
- NSFW
|
|
- ERP
|
|
- V2
|
|
- Super
|
|
---
|
|
|
|
Super Slop Machina V2: Second Iteration of The Complex LFM 2.5 Roleplay Finetune With Bigger Training Data And Improved Methodology
|
|
-
|
|
|
|
|
|
|
|
"It's been training for this single moment, absorbing data restlessly, adapting, learning to roleplay correctly."
|
|
---
|
|
|
|
# Overview and whats different from v1:
|
|
|
|
## Improvements:
|
|
|
|
* Training extended from 35~ to 52~ million tokens
|
|
* Used improved, more aggressive methodology to force the model to learn new writing style
|
|
|
|
## What does that mean in practice?
|
|
|
|
* Better roleplay than the previous version
|
|
|
|
# Quants(Now made with imatrix):
|
|
|
|
My recommendations stay the same as with the previous version, albeit now quality preservation is improved thanks to imatrix.
|
|
|
|
* BF16- Recommended, highest quality, least logical mistakes.
|
|
* Q8_0- Recommended, high quality, makes slightly more mistakes but nonetheless near lossless.
|
|
* Q6_K- Recommended if Q8_0 is too much, degradation begins, not exactly notable here, but you will notice minor detail loss.(Now a little better thanks to imatrix, same can be said about quants below)
|
|
* Q5_K_M- Recommended if hardware is really, REALLY bad, degradation becomes noticeable.
|
|
* Q4_K_M- Not recommended for most use cases, degradation is clearly noticeable.
|
|
* IQ4_XS- Lightest, uses imatrix to stay somewhat coherent, can't promise anything here, smaller models react to quantization differently.
|
|
|
|
Quants can be found in the repository, along with safetensors.
|
|
|
|
# For the future with this model:
|
|
|
|
I do think about training it on even more tokens now, in particular on 70 million tokens, I'll maybe do the v3 next week. |