Skip to content
3D

LFM3-VL-2B 3D Coder

SFT для генерации Three.js-сцен

1 min readupdated

Экспериментальный SFT-файнтюн LiquidAI/LFM3-VL-2B от AlekseyCalvin: по текстовой инструкции генерирует исполняемые Three.js-сцены и ассеты, экспортируемые в GLB. Это квази-воспроизведение исследования ccore/Qwen3.5-4B-3D-Coder; обучение ещё идёт, итоговые метрики в карточке не опубликованы.

Ключевые особенности

  • База — LiquidAI/LFM3-VL-2B; QLoRA: bitsandbytes NF4 + double quant, LoRA r=64, α=128, 65 874 944 обучаемых параметров (3.32%).
  • Vision encoder заморожен; LoRA на языковые проекции q/k/v/o/gate/in_proj/out_proj/w1/w2/w3.
  • Данные: ccore/ThreeJS-3D-Scenes и 1 500 примеров из ccore/ThreeJS-3D-GLB-Assets; примеры длиннее 24 576 токенов исключены.
  • Каждые 200 шагов blocking HTML-emission gate: 0/2 валидных HTML останавливает запуск; предыдущий DFT-run коллапсировал на шаге 200.
  • Обучение на одной NVIDIA L4 (24 GB) в Google Colab, effective batch 4, peak LR 2e-5, 1 эпоха.

Пример

from transformers import AutoModelForImageTextToText, AutoProcessor
import torch

model_id = 'AlekseyCalvin/LFM3-VL-2B-3D-Coder-SFT'
model = AutoModelForImageTextToText.from_pretrained(
    model_id, dtype=torch.bfloat16, device_map='auto', trust_remote_code=True)
processor = AutoProcessor.from_pretrained(model_id, trust_remote_code=True)
ПодробнееПодробнее
Результаты: TBD — запуск ещё идёт; кривые в metrics.jsonl, сэмплы в evals/step-*, вердикты gate в gen_gate_*.json.
Железо: одна NVIDIA L4 (24 GB) в Google Colab, micro-batch 1, grad-accum 4.
Ограничение: примеры длиннее 24 576 токенов исключены; merged, final_adapter и GGUF появятся по завершении.
Метод: TRL SFTTrainer, loss_type='nll', assistant-only loss, ChatML с <think>.
Discuss Telegram 0

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 3.8 GB · ready files4.7 GB
  • FP8 / INT8recommendedweights 2.0 GB2.7 GB
  • 4-bitweights 1.3 GB1.9 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.

This is a LoRA: counted together with its base model LFM3-VL-2B.

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.

    # AlekseyCalvin/LFM3-VL-2B-3D-Coder-SFT
    # ===== Tokens: paste yours between the quotes =====
    export HF_TOKEN=""        # huggingface.co/settings/tokens, for gated models
    # ==========================================
    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 LiquidAI/LFM3-VL-2B --local-dir ~/models/LFM3-VL-2B
    hf download AlekseyCalvin/LFM3-VL-2B-3D-Coder-SFT --local-dir ~/models/LFM3-VL-2B-3D-Coder-SFT
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to LFM3-VL-2B
  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

Requires the base model

LiquidAI/LFM3-VL-2B

This model is a LoRA / adapter. Download the base model first, then load this one on top.

  1. LFM3-1.7B-Basefamily root
  2. LFM3-VL-2B
  3. LFM3-VL-2B-3D-Coder-SFT

New posts about versions of LFM3-1.7B-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 9All on Hugging Face

About the model

AlekseyCalvin/LFM3-VL-2B-3D-Coder-SFT
Author
AlekseyCalvin
Category
3D
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

3D

EcoVGGT

EcoVGGT