Gemma 4 E4B it SDFT Heretic RP GGUF
This repository contains GGUF quantized files for the main merged model:
Ilya626/gemma-4-E4B-it-SDFT_Heretic_RP
These are inference-ready GGUF artifacts for llama.cpp-compatible runtimes. The full merged Hugging Face weights are published separately in the base model repository above.
Files
- gemmae4b_lr5e6_resume_r8-Q3_K_L.gguf - lower memory option.
- gemmae4b_lr5e6_resume_r8-Q4_K_M.gguf - recommended default.
- gemmae4b_lr5e6_resume_r8-Q5_K_M.gguf - higher quality option.
- gemmae4b_lr5e6_resume_r8-Q6_K.gguf - best quality GGUF option here.
- gemmae4b_lr5e6_resume_r8-mmproj-f16.gguf - shared multimodal projector file.
Notes
This is an RP-enhanced model based on the Heretic SDFT line. It is intended for research, evaluation, local RP testing, and narrative experiments.
The underlying model may have weakened safety guardrails compared with standard aligned chat models. Do not use it to generate harmful, illegal, abusive, exploitative, or otherwise unsafe content.
Suggested Starting Settings
temperature: 0.8-0.95
top_p: 0.95
top_k: 64
For most users, start with Q4_K_M. Use Q5_K_M or Q6_K if you have the memory and want better prose stability.
