base_model:
- LiquidAI/LFM2-8B-A1B license: apache-2.0 language:
- en pipeline_tag: text-generation library_name: transformers tags:
- finetune
- unsloth
- mixture of experts
- sparse moe
- moe
- heretic
- uncensored
- aliterated
- finetune
- unsloth
- finetune
- all use cases
- coder
- creative
- creative writing
- fiction writing
- plot generation
- sub-plot generation
- fiction writing
- story generation
- scene continue
- storytelling
- fiction story
- science fiction
- romance
- all genres
- story
- writing
- vivid prosing
- vivid writing
- fiction
- roleplaying
- bfloat16
- all use cases
Fine tune of EXPANDED "LFM2-8B-A1B" to "LFM2-12B-A1B" (almost 50%, custom scripting) using Unsloth using custom dataset(s), 128k context in 16 bit precision.
Expands base model from 24 to 32 layers, 256 to 342 tensors to give the model both more knowledge and brainpower.
Trained on Deckard II [5 datasets, in house]:
Excels at long form creative generation. Very intelligent too.
This is full uncensored version (uncensored first, trained second)
This model is a sparse mixture of experts model (32) with 4 experts activated.
Speed exceeds 50-100 t/s on CPU // 200 t/s on most cards // 400 t/s + on 5090 at QUANT Q6K [4 experts].
One example generation below.
Can also be used on phones // mobile devices.
IN HOUSE BENCHMARKS [by Nightmedia]:
arc-c arc/e boolq hswag obkqa piqa wino
LFM2-12B-A1B-SpeedDemon-The-Deckard-II-HERETIC-Uncensored
bf16 0.469,0.612,0.782,0.676,0.412,0.743,0.610
---
BASE UNTUNED MODEL:
LFM2-8B-A1B
bf16 0.464,0.583,0.826,0.624,0.398,0.717,0.575
mxfp8 0.460,0.575,0.829,0.624,0.394,0.711,0.567