Skip to content
Other

Ternary-Flint-40B на Intel Arc

100+ tok/s через OpenCL

2 min readupdated

Это репозиторий с OpenCL-рантаймом для ternary MoE-модели Ternary-Flint-40B в GGUF PQ2_0. Код на Python (numpy + pyopencl) запускает 2-битную модель на одной Intel Arc A770 16 ГБ. Автор добавляет только рантайм: сам GGUF — неизменённая копия кванта от kabcpc, базовая модель Flint — от RecursiveSI.

Ключевые особенности

  • 106,3 tok/s на коротком контексте и 102,9 tok/s в среднем до 2048 позиций на одной A770 16 ГБ.
  • Минимальный OpenCL-рантайм: один Python-файл на numpy и pyopencl, без тяжёлого стека.
  • В репозитории лежат GGUF PQ2_0 на ~11 ГБ, tokenizer.json и код; квантование сделал kabcpc.
  • Есть команды check, speed, prompts и long для проверки корректности, скорости и perplexity.
  • Пока только batch 1 и greedy-декодирование; автор ищет тестеры на других Intel Arc.

Пример

pip install -r requirements.txt
python flint_rt.py check --gguf Ternary-Flint-40B-PQ2_0-Calibrated-11GB.gguf --tokenizer tokenizer.json
python flint_rt.py speed --gguf Ternary-Flint-40B-PQ2_0-Calibrated-11GB.gguf --tokenizer tokenizer.json
ПодробнееСкорость: 106,3 tok/s на коротком контексте, 102,9 tok/s в среднем до позиции 2048.
Prompt processing: 113 tok/s для 512 токенов (4,52 с); загрузка из кэша ~8 с.
Perplexity: 13,38 на 1600 токенах WikiText-2 (top-1 56,0%).
Нужны Intel Arc A770 16 ГБ, драйвер с OpenCL и расширениями cl_intel_subgroups, Python 3.10+, numpy, pyopencl, tokenizers; 10,3 ГБ на кэш.
Ограничения: batch 1, только greedy, без sampling, chat template и сервера; prefill по одному токену; тестировалось только на A770 под Windows.
Discuss Telegram 0 4

Will it run?

Runs in GGUF Q2_0

The most precise of the ready quantised files fits.

  • GGUF Q2_0weights 11 GB · ready files13 GB

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: GGUF Q2_0 — from RTX 5080

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.

    # TheSanchezian/Ternary-Flint-40B-Intel-Arc
    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 TheSanchezian/Ternary-Flint-40B-Intel-Arc --local-dir ~/models/Ternary-Flint-40B-Intel-Arc
    # 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

New posts about versions of Ternary-Flint-40B-Intel-Arc (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

TheSanchezian/Ternary-Flint-40B-Intel-Arc
Author
TheSanchezian
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

40

Ask AI about this model

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

Comments

Language models

Flint

Flint