Skip to content
Computer vision

Groot-19B-A3B-261023

меньше параметров, выше средний балл

2 min read

Groot-19B-A3B-261023 — нативно мультимодальная MoE-модель от GrootTeam для сложных reasoning- и tool-задач. У неё ~19B общих параметров, но активируется ~3B на токен; это самая маленькая модель в своей группе сравнения. Контекст — 262 144 токена нативно, до ~1M с YaRN.

Ключевые особенности

  • 40 слоёв: 10 full-attention и 30 linear-attention; 128 routed + 1 shared expert, 8 routed активны на токен.
  • Релизы BF16, FP8, NVFP4 и GGUF Q4_K_M; BF16 — ~39 ГБ, GGUF Q4_K_M — 11.7 ГБ.
  • Обходит Qwen3.6-35B-A3B и Gemma-4-26B-A4B на 12 из 16 бенчмарков и отвечает короче.
  • Thinking mode по умолчанию, поддержка tool calling и MTP-слой на 0.4B.
  • Запуск через Transformers ≥5.8.1, vLLM ≥0.19.1, SGLang ≥0.5.10 и KTransformers.

Пример

from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "Groot/Groot-19B-A3B-261023"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name, dtype="auto", device_map="auto")

messages = [{"role": "user", "content": "Give me a short introduction to large language models."}]
ПодробнееБенчмарки: Terminal Bench 2.1 — 52.10 против 43.82 у Qwen3.6-35B-A3B и 26.52 у Gemma-4-26B-A4B; средний балл по 16 тестам — 41.44 против 38.08 и 30.05.
Железо: BF16 (~39 ГБ) — одна 48/80-ГБ GPU; FP8/NVFP4 — полный 256K на RTX 5090; GGUF Q4_K_M — 128K на 16 ГБ, 256K на 24 ГБ.
Обучение: старт с Qwen3.6-35B-A3B и пост-тренинг на менее чем 0.1B токенов open-source данных.
Ограничение: YaRN в open-source рантаймах статический и может слегка ухудшать качество на обычных длинах — включайте только при необходимости.
Примеры
Discuss Telegram 81 959

Will it run?

Runs in FP8 / INT8

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

  • BF16weights 36 GB · ready files40 GB
  • FP8 / INT8recommendedweights 18 GB20 GB
  • 4-bitweights 10 GB12 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 4090

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.

    # GrootTeam/Groot-19B-A3B-261023
    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 GrootTeam/Groot-19B-A3B-261023 --local-dir ~/models/Groot-19B-A3B-261023
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to Groot-19B-A3B-261023
  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 Groot-19B-A3B-261023 (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face

About the model

GrootTeam/Groot-19B-A3B-261023
Author
GrootTeam
Category
Computer vision
Released
Posted

On NeyroLearn

Hugging Face

95981

Ask AI about this model

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

Comments

Robotics

GR00T N1.7

GR00T N1.7