Skip to content
Language models

Supra2.5-Pro-Instruct

ultra-compact 65M parameter LLM

2 min readupdated

Supra2.5-Pro-Instruct is a tiny English language model from SupraLabs, fine-tuned for conversation and basic instruction following. Built on the Supra2.5-Pro-Base foundation model, it has just 65 million parameters (roughly 0.07B). Unlike larger models, this release focuses on ultra-lightweight local inference (running the model directly on your device) and supports the standard ChatML chat template. It is well-suited for developers and enthusiasts experimenting with ultra-small models or running AI on low-end hardware.

Key features

  • Minimal footprint: At just 65M parameters, it runs locally with minimal resource overhead.
  • Trained on synthetic data: The model underwent 2 epochs of fine-tuning on nearly 9,000 synthetic conversations generated by DeepSeek V4.1 Flash 1017 using prompts from SmolTalk2.
  • Ready-to-use quantizations: Alongside the original weights, optimized GGUF formats (F16, Q8_0, and Q4_K_M) are available for fast execution via llama.cpp and compatible clients.
  • Uncensored and unaligned: The authors note that the model has not undergone safety alignment or preference optimization.

What you can do

  • Run a local chatbot on a low-end PC: Use the GGUF Q4_K_M or Q8_0 builds for simple, everyday English chats without needing a GPU.
  • Test basic code and math: Use the model as an ultra-fast drafting tool for elementary coding tasks and simple calculations within its capabilities.
  • Embed into a custom pipeline: Integrate it as a lightweight ChatML-compatible component, provided you add an external safety filter before displaying outputs to users.

Example

This sample conversation from the model card highlights both its conversational style and its tendency to hallucinate facts due to its small size:
User: How are you?
Supra: I'm running well, thanks for asking! As a compact language model, I don't have feelings or memory across separate chats. What can I help you with today?

Read moreDetails
Parameters: 65M (0.07B), English language.
Training: 2 epochs on ~9,000 dialogues from DeepSeek V4.1 Flash 1017 (using the SmolTalk2 prompt base).
Formats and runtime: Compatible with transformers and the ChatML template; available in GGUF F16, Q8_0, and Q4_K_M.
Limitations: No safety alignment; prone to factual errors due to its small size (such as inaccuracies in Einstein's biography in the model card examples).
Discuss Telegram 19 864

Will it run?

Runs in F32

Fits without quantisation.

  • F32weights 0.2 GB · ready files1.1 GB
  • GGUF F16weights 0.1 GB · ready files1.0 GB
  • BF16weights 0.1 GB1.0 GB
  • GGUF Q8_0recommendedweights 0.1 GB · ready files0.9 GB
  • FP8 / INT8weights 0.1 GB0.9 GB
  • GGUF Q4_K_Mweights 0.1 GB · ready files0.9 GB
  • 4-bitweights 0.0 GB0.9 GB

Tip: go 8-bit.GGUF Q8_0 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: GGUF Q8_0 — from RTX 2060 · F32 — from RTX 2060

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

Language models: with a 8192-token context.

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

Run it on Massed Compute

A ready script for the machine we advise above: paste it and the model downloads and starts.

  1. 1. Rent A30

    $0.26 an hour with the coupon. Pick the Ubuntu image when you create the machine.

    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.

    # SupraLabs/Supra2.5-Pro-Instruct — A30, GGUF Q8_0
    # NVIDIA driver: CUDA 13 needs 580+. If older or missing, the current nvidia-open is installed (keep yours: KEEP_DRIVER=1)
    DRV=$(nvidia-smi --query-gpu=driver_version --format=csv,noheader 2>/dev/null | head -1 | cut -d. -f1)
    case "$DRV" in ""|*[!0-9]*) DRV=0 ;; esac
    if [ "$DRV" -lt 580 ] && [ -z "${KEEP_DRIVER:-}" ]; then
      . /etc/os-release && \
      curl -fsSLO "https://developer.download.nvidia.com/compute/cuda/repos/ubuntu${VERSION_ID//./}/x86_64/cuda-keyring_1.1-1_all.deb" && \
      sudo dpkg -i cuda-keyring_1.1-1_all.deb && rm -f cuda-keyring_1.1-1_all.deb && \
      sudo apt-get update && \
      sudo DEBIAN_FRONTEND=noninteractive apt-get install -y nvidia-open || \
      { echo "Could not install the NVIDIA driver: see the error above. To keep the current one, run with KEEP_DRIVER=1."; exit 1; }
      echo "NVIDIA driver updated. Reboot the machine (sudo reboot) and run this same script again."
      exit 0
    fi
    curl -fsSL https://ollama.com/install.sh | sh
    ollama run hf.co/SupraLabs/Supra2.5-Pro-Instruct:Q8_0
    Add more LoRAs to Supra2.5-Pro-Instruct
  3. 3. Use it

    The chat opens right in the terminal. Leave with /bye, start again with the same ollama run command.

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.

  • MTP (multi-token prediction)0.34.0from November 16, 2026new release

    1.5–2× generation speed

    If the model has MTP layers (DeepSeek-V3/R1, Qwen3-Next, GLM, MiMo and others), it predicts several tokens per step itself.

    How to enable: vLLM / SGLang: --speculative-config '{"method": "mtp", "num_speculative_tokens": 1}'. Check the model card.

    Earlier generations (2)
    • EAGLE-3 — Speculative decoding with a small draft head. For models without their own MTP layers.
    • EAGLE-2 / Medusa — Earlier draft-head generations. Still work if that is all you find.
  • FP8 KV cache0.34.0from November 16, 2026new release

    2× more context in the same memory

    Keeps the attention cache in 8 bits, so long contexts fit where they did not.

    How to enable: vLLM: --kv-cache-dtype fp8; llama.cpp: -ctk q8_0 -ctv q8_0.

  • vLLM / SGLang0.34.0from November 16, 2026new release

    many times more tokens/s on a server

    Server engines: batching, FlashAttention, FP8 and speculative decoding out of the box.

    How to enable: vllm serve <model> — an OpenAI-compatible API.

    Earlier generations (1)
    • llama.cpp — GGUF models on any GPU, Mac or even CPU; some layers can stay in RAM.

Model family

New posts about versions of Supra2.5-Pro-Instruct (quants, finetunes, LoRAs) will show up under the bell.

No known relatives of this model yet. All derivatives on Hugging Face

About the model

SupraLabs/Supra2.5-Pro-Instruct
Author
SupraLabs
Category
Language models
Released
Posted

On NeyroLearn

Hugging Face

86419

Ask AI about this model

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

Comments