Instructions to use sh0ck0r/heretic-rp-vision-31b with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use sh0ck0r/heretic-rp-vision-31b with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("image-text-to-text", model="sh0ck0r/heretic-rp-vision-31b") messages = [ { "role": "user", "content": [ {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"}, {"type": "text", "text": "What animal is on the candy?"} ] }, ] pipe(text=messages)# Load model directly from transformers import AutoProcessor, AutoModelForMultimodalLM processor = AutoProcessor.from_pretrained("sh0ck0r/heretic-rp-vision-31b") model = AutoModelForMultimodalLM.from_pretrained("sh0ck0r/heretic-rp-vision-31b") messages = [ { "role": "user", "content": [ {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"}, {"type": "text", "text": "What animal is on the candy?"} ] }, ] inputs = processor.apply_chat_template( messages, add_generation_prompt=True, tokenize=True, return_dict=True, return_tensors="pt", ).to(model.device) outputs = model.generate(**inputs, max_new_tokens=40) print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:])) - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- vLLM
How to use sh0ck0r/heretic-rp-vision-31b with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "sh0ck0r/heretic-rp-vision-31b" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "sh0ck0r/heretic-rp-vision-31b", "messages": [ { "role": "user", "content": [ { "type": "text", "text": "Describe this image in one sentence." }, { "type": "image_url", "image_url": { "url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg" } } ] } ] }'Use Docker
docker model run hf.co/sh0ck0r/heretic-rp-vision-31b
- SGLang
How to use sh0ck0r/heretic-rp-vision-31b with SGLang:
Install from pip and serve model
# Install SGLang from pip: pip install sglang # Start the SGLang server: python3 -m sglang.launch_server \ --model-path "sh0ck0r/heretic-rp-vision-31b" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "sh0ck0r/heretic-rp-vision-31b", "messages": [ { "role": "user", "content": [ { "type": "text", "text": "Describe this image in one sentence." }, { "type": "image_url", "image_url": { "url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg" } } ] } ] }'Use Docker images
docker run --gpus all \ --shm-size 32g \ -p 30000:30000 \ -v ~/.cache/huggingface:/root/.cache/huggingface \ --env "HF_TOKEN=<secret>" \ --ipc=host \ lmsysorg/sglang:latest \ python3 -m sglang.launch_server \ --model-path "sh0ck0r/heretic-rp-vision-31b" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "sh0ck0r/heretic-rp-vision-31b", "messages": [ { "role": "user", "content": [ { "type": "text", "text": "Describe this image in one sentence." }, { "type": "image_url", "image_url": { "url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg" } } ] } ] }' - Docker Model Runner
How to use sh0ck0r/heretic-rp-vision-31b with Docker Model Runner:
docker model run hf.co/sh0ck0r/heretic-rp-vision-31b
Heretic RP Vision 31B
Heretic RP Vision 31B is a roleplay and creative-writing fine-tune from the Heretic Gemma 4 31B line. It was trained to improve anime-style character interaction, long-form roleplay flow, emotionally driven dialogue, and character card generation while preserving the model's multimodal stack.
The merged release keeps the Gemma 4 conditional-generation architecture and processor files intact, including the vision processor. In local testing, the multimodal behavior remained functional after the RP fine-tune.
Intended Use
- Anime and character roleplay
- Long-form interactive fiction
- Persona-driven chat
- Creative writing and scene continuation
- Character profile and first-message generation
- Multimodal chat where image understanding is useful
Training Summary
This model was trained with QLoRA and merged back into the base model for this release.
| Setting | Value |
|---|---|
| Model size | 31B |
| Fine-tuning method | QLoRA |
| Training steps | 2500 |
| Final epoch | 1.287 |
| Context length | 4096 |
| LoRA rank | 16 |
| LoRA alpha | 32 |
| LoRA dropout | 0.05 |
| Peak learning rate | 1e-4 |
| Scheduler | Cosine |
| Precision | bf16 |
| Final logged loss | 1.577 |
The training mix contained about 31k examples from roleplay, anime character dialogue, creative writing, prose, and instruction-style character generation sources. One encrypted dataset archive was excluded from the active mix because it could not be read by the dataset loader.
Behavior
Compared with the base Heretic Gemma 4 31B model, this release should lean more strongly into:
- Maintaining character voice across longer scenes
- Writing detailed first messages and roleplay openings
- Creating emotionally charged character dynamics
- Supporting rivals, dangerous allies, bonded companions, cursed partners, and supernatural relationship hooks
- Continuing scenes with more dramatic anime-style pacing
Example Prompts
Character generation:
Create an original anime roleplay character built around an unstable emotional
bond with the user. Include appearance, personality, backstory, current
situation, relationship hook, secrets or complications, and a first in-character
message.
Scene continuation:
Stay in character and continue the scene with detailed narration, emotional
tension, and natural dialogue. Keep the conflict alive instead of resolving it
too quickly.
Dangerous ally setup:
Create a dangerous ally character who saves or shelters the user, but has their
own agenda. The user should need them, while trusting them remains risky.
Limitations
This is an experimental creative model. It may produce inconsistent lore, heightened melodrama, familiar anime tropes, or overly intense relationship dynamics depending on prompting. Users should validate multimodal behavior in their own inference stack, especially if using nonstandard loaders.
Notes
This release is intended for creative and roleplay workflows. The model may reflect style and content patterns from its training mix, including dramatic fiction, supernatural themes, conflict-heavy dialogue, and anime-inspired character tropes.
Built by sh0ck0r as part of the Heretic Gemma 4 experimentation line.
- Downloads last month
- 18
