Skip to content
Other

Charlie320/zils-mini-m15p

a compact fine-tune for Bittensor subnet 127

2 min readupdated

This model is a fine-tuned version of Qwen/Qwen3.5-4B (an open-weight language model with 4.54 billion parameters) built for the Zils d3-mini ecosystem within Bittensor subnet 127. It was created by averaging two rank-64 LoRA adapters (small add-on layers used to quickly adapt a base model) trained with different seeds on specialized datasets. The model retains vision capabilities from its parent configuration, was trained in BF16 (a reduced-precision floating-point format), and is geared toward specialized decision-making and text tasks.

Key features

  • Built on the modern Qwen/Qwen3.5-4B model while preserving the vision tower from the parent model.
  • Trained on a 61,091-row dataset containing exam-style problems and structured solutions.
  • Created by averaging the weights of two LoRA adapters with equal weighting.
  • Uses soft training targets to balance model responses with base probabilities.

What you can do

  • Deploy the model as a decision-making component within the Bittensor subnet 127 ecosystem.
  • Run local inference of this 4.54-billion-parameter model in BF16 for text-and-vision tasks.
  • Evaluate how the model handles exam-style formats and classification tasks.
  • Test function calling capabilities enabled by training on the glaive-function-calling-v2 dataset.

Example

The model does not include bundled inference scripts, so you can load the weights the standard way through the Transformers library using the base Qwen3.5 architecture.

Read moreDetails
Base model: Qwen/Qwen3.5-4B and joamCriff123/zils-mini-m1115
Parameter count: 4.54 billion
Weight format: BF16
Training method: averaging of two LoRAs (rank 64) at a 5e-5 learning rate
Discuss Telegram 0

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 8.5 GB · ready files9.8 GB
  • FP8 / INT8recommendedweights 4.2 GB5.2 GB
  • 4-bitweights 2.4 GB3.1 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.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: FP8 / INT8 — from RTX 2060 · BF16 — from RTX 3080

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

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.

    # Charlie320/zils-mini-m15p
    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 Charlie320/zils-mini-m15p --local-dir ~/models/zils-mini-m15p
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to zils-mini-m15p
  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. Qwen3.5-4B-Basefamily root
  2. Qwen3.5-4B
  3. zils-mini-m15p

New posts about versions of Qwen3.5-4B-Base (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 12All on Hugging Face
Merges 1All on Hugging Face

About the model

Charlie320/zils-mini-m15p
Author
Charlie320
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

00

Ask AI about this model

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

Comments