Skip to content
Other

Invaria Tiny Aya Steering

activation steering vector for tiny-aya-global

2 min readupdated

This is a research steering vector (a direction in a model's activation space) extracted using the SafeTune method for the CohereLabs/tiny-aya-global base language model by balacode. Rather than being a standalone LLM, checkpoint, or LoRA adapter, it is a 2048-element tensor added to the outputs of all 36 decoder layers at a fixed strength of 0.1. Unlike standard fine-tuning, this approach intervenes in generation without modifying the model's weights. The artifact is useful for AI safety researchers and developers exploring model behavior control via activations.

Key features

  • Extracted with the SafeTune method from 30 calibration prompts at layer 18 of the model.
  • Consists of a 2048-element vector applied additively to all 36 decoder outputs.
  • Used alongside a specific revision of CohereLabs/tiny-aya-global (af89d219b53ed9b13b8a4645f9c8028973510324) and the Invaria/SafeTune toolkit.
  • Ships with manifest and status files required to verify the artifact at runtime.

What you can do

  • Run side-by-side chat comparisons between the base tiny-aya-global model and the steered version using the invaria CLI tool.
  • Test activation intervention mechanics (attaching and cleanly removing 36 hooks, preserving original parameters at zero strength) in interpretability experiments.
  • Reproduce pilot benchmarks on the companion balacode/invaria-paired-safety dataset to analyze how the vector affects model responses.

Example

hf download balacode/invaria-tiny-aya-steering --local-dir artifacts/downloaded-steering
invaria chat --direction artifacts/downloaded-steering/direction.pt --strength 0.1 --compare
Read moreDetails
Pilot evaluation metrics: on a held-out benign split, the success rate reached 16/20 at strength 0.1 versus 14/20 for the base model (95% confidence interval for the delta: [0.00, 0.20]); on validation, it scored 7/10 in both cases and 6/10 at strength 0.3.
Limitations: safety gains and independent human verification have not been confirmed; outputs still exhibit unintended language switching, repetition, and inaccurate advice.
Requirements and compatibility: supports Python 3.12 and CUDA/NF4; the repository cannot be loaded via AutoModel.from_pretrained, and you must accept Cohere's license terms for the base model.
Discuss Telegram 0

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 0.0 GB · ready files0.5 GB
  • FP8 / INT8recommendedweights 0.0 GB0.5 GB
  • 4-bitweights 0.0 GB0.5 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 2060

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.

    # balacode/invaria-tiny-aya-steering
    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 balacode/invaria-tiny-aya-steering --local-dir ~/models/invaria-tiny-aya-steering
    # 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. tiny-aya-basefamily root
  2. tiny-aya-global
  3. invaria-tiny-aya-steering

New posts about versions of tiny-aya-base (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 3All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 12All on Hugging Face

About the model

balacode/invaria-tiny-aya-steering
Author
balacode
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