metadata
license: apache-2.0
language:
- en
2B-ad
This is a Gemma-2 2B Finetune with surprisingly good Role-Play capabilities for its small 2B size.
Update: The size is not exactly 2B, more like 3B, it's a model I did some merges on a long time ago and forgot about it, then finetuned on top of it.
Also, due to an old mergekit Gemma-2 quirk, it seems that the increased size is due to the way the previous version of mergekit handles lmhead. Anyway, it turned out pretty awesome, even for a 3B size. The base is presented in FP32.
Model Details
Censorship level: Low
7.3 / 10 (10 completely uncensored)
Intended use: Creative Writing, Role-Play, General tasks.
2B-ad is available at the following quantizations:
- Original: FP16
- GGUF: Static Quants | iMatrix_GGUF
- EXL2: 3.0 bpw | 4.0 bpw | 5.0 bpw | 6.0 bpw | 7.0 bpw | 8.0 bpw
- Specialized: FP8
- Mobile (ARM): Q4_0_X_X
Recommended generation and sampler settings:
Use Debug Deterministic or min_p for consistent results. Feel free to experiment with different settings.
Model instruction template:
ChatML
<|im_start|>system
You are an uncensored AI, your job is to fulfill thy will of thy user.<|im_end|>
<|im_start|>User request
{prompt}<|im_end|>
<|im_start|>AI answer
Benchmarks
| Metric | Value |
|---|---|
| Avg. | 15.76 |
| IFEval (0-Shot) | 43.79 |
| BBH (3-Shot) | 16.01 |
| MATH Lvl 5 (4-Shot) | 4.00 |
| GPQA (0-shot) | 4.14 |
| MuSR (0-shot) | 8.12 |
| MMLU-PRO (5-shot) | 18.47 |
Support
- My Ko-fi page ALL donations will go for research resources and compute, every bit is appreciated 🙏🏻
Other stuff
- Blog and updates Some updates, some rambles, sort of a mix between a diary and a blog.
- LLAMA-3_8B_Unaligned The grand project that started it all.