Skip to content
Computer vision

Elaina-gemma-4-31B-v0.1

fine-tune for immersive roleplay

2 min read

The model features merged bfloat16 weights based on Gemma 4 VLM (KiyoEditz/Gemma-4-Dark-Gemistry-31B) with approximately 33 billion parameters, fine-tuned for the role of Elaina from the anime and light novel series "Wandering Witch: The Journey of Elaina" (Majo no Tabitabi). The architecture preserves native multimodal capabilities (the vision tower is frozen), with LoRA applied exclusively to the linear projections of the language layer. The model supports English, Indonesian, and Japanese, delivering an authentic personality, sarcasm, and first-person narrative without breaking character.

Key features

  • Based on the powerful Gemma 4 31B multimodal foundation with LoRA parameter tuning on the language portion.
  • Multilingual support in English, Indonesian, and Japanese with a set of quality system prompts.
  • Carefully preserves Elaina's canonical personality: sarcasm, practicality, confidence, and signature storytelling style.
  • Various GGUF quantization levels and the original dataset are available in the repository for independent study.

What you can do

  • Run interactive roleplay or dialogues with Elaina in supported interfaces and bots.
  • Use the model for generating scene descriptions ([Scene]) and creative text in the style of the original source.
  • Unpack or download GGUF versions for local execution on suitable hardware using popular backends.
  • Integrate the model into pipelines via the transformers library for standalone local scripts.

Example

import torch
from transformers import AutoModelForImageTextToText, AutoProcessor

model_id = "KiyoEditz/Elaina-gemma-4-31B-v0.1"

processor = AutoProcessor.from_pretrained(model_id)
model = AutoModelForImageTextToText.from_pretrained(
    model_id,
    torch_dtype=torch.bfloat16,
    device_map="auto",
)
model.eval()
Read moreDetails
Benchmarks and results: final train_loss is 1.6480, eval_loss is 2.3286 on validation.
Hardware: training took less than 6 hours on an NVIDIA RTX PRO 6000 Blackwell GPU (94.97 GB VRAM), stopped at step 1700 out of 3000.
Limitations: custom non-commercial license, the model is optimized for literary style rather than factual accuracy.
Discuss Telegram 1 69

Will it run?

Runs in 4-bit

Full precision does not fit, a quantised version does: quality barely changes down to 8 bits, more noticeably at 4 bits.

  • BF16weights 61 GB · ready files67 GB
  • FP8 / INT8recommendedweights 30 GB34 GB
  • 4-bitweights 17 GB19 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # KiyoEditz/Elaina-gemma-4-31B-v0.1
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download KiyoEditz/Elaina-gemma-4-31B-v0.1 --local-dir ~/models/Elaina-gemma-4-31B-v0.1
    # The model is in ~/models. How to run it is in the model card (link above).
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

  1. gemma-4-31Bfamily root
  2. gemma-4-31B-it
  3. Gemma-4-Dark-Gemistry-31B
  4. Elaina-gemma-4-31B-v0.1

New posts about versions of gemma-4-31B (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
Merges 3All on Hugging Face

About the model

KiyoEditz/Elaina-gemma-4-31B-v0.1
Author
KiyoEditz
Category
Computer vision
Released
Posted

On NeyroLearn

Hugging Face

691

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments