Downloads · 30 days
13
23% of all-time downloads
gemma-4-31B-it-heretic
Vampelium/gemma-4-31B-it-heretic
gemma-4-31B-it-heretic is a image-text-to-text model from Vampelium. Use it for the image-text-to-text task on the model card, and read the license before you ship it in a product. It is set up for transformers. The card lists the license as gemma.
Vampelium's local Gemma 4 31B IT checkpoint after Heretic abliteration.
All-time downloads
57
Public
Parameters
31.3B
62.6 GB on disk
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors62.5 GB · 100%
At a glance
- Task
- Image-Text-to-Text
- Library
- transformers
- License
- gemma
- Model type
- gemma4
- Access
- Public
- Created
- Aug 30, 2026
- Updated
- Aug 30, 2026
- SHA
- 88a86a0b
Try a prompt
Base models
Model card
From the Hugging Face model README
gemma-4-31B-it-heretic
Vampelium's local Gemma 4 31B IT checkpoint after Heretic abliteration.
Produced on 4x NVIDIA A100.
Shards were reassembled from Google Drive split-download names
(model-0000N-of-00013-00X.safetensors) back to the filenames in
model.safetensors.index.json. All 1188 tensors match the index; payload
size is 62,546,336,448 bytes (~58.3 GiB / 62.5 GB BF16).
Files
| File | Role |
|---|---|
model-00001-of-00013.safetensors … model-00013-of-00013.safetensors | BF16 weights |
model.safetensors.index.json | Shard map |
config.json / generation_config.json | Model + decode settings |
tokenizer.json / tokenizer_config.json / chat_template.jinja | Tokenizer + chat template |
processor_config.json | Image / audio / video processor |
Load
This checkpoint was saved with transformers 5.6.2. Gemma 4 needs a recent
transformers that includes Gemma4ForConditionalGeneration.
from transformers import AutoTokenizer, AutoModelForImageTextToText
model_id = "Vampelium/gemma-4-31B-it-heretic"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForImageTextToText.from_pretrained(
model_id, dtype="auto", device_map="auto"
)
Full BF16 needs ~62 GB of GPU memory. A single 24 GB / 48 GB card will not
load this unquantized — use device_map="auto" across multiple GPUs, 4-bit
(bitsandbytes), or a GGUF quant for llama.cpp / Ollama.
Gemma 4 is a hybrid thinking model. Abliteration is aimed at the default
non-thinking path (enable_thinking=False). Use that for roleplay and
creative writing; if you enable thinking, raise max_new_tokens so the
final answer is not cut off.
Disclaimer
Safety alignment was removed. You are responsible for how you use this model and for following applicable law. The base Gemma license still applies.
- Task
- Image-Text-to-Text
- Library
- transformers
- Type
- gemma4
- License
- gemma
- Created
- Aug 30, 2026
- Updated
- Aug 30, 2026