Skip to content
Language models

LLaDA2.2-mini-Thinking: A reasoning model for block-diffusion MoE

2 min read

This model is an upgraded version of LLaDA2.2-mini (a 16B block-diffusion MoE ( Mixture of Experts ) with 1.4B active parameters), fine-tuned to reason inside … tags before generating a response or calling tools. The base model answers immediately without reasoning in text. The new version is trained on agent sessions and reasoning examples from DeepSeek, combining generation capabilities with logical task analysis.

Key features

  • Added reasoning before the response: the model explicitly writes out its thought process in tags.
  • Retained native support for tool calls from the base architecture.
  • Trained on 2,003 agent task sessions and 3,000 reasoning examples (25.6M tokens total).
  • Weights merged in bf16 format, training took 5.3 hours on a single A100 80GB GPU.

What you can do

  • Run a local server via SGLang with a deepseek-r1 style reasoning parser.
  • Integrate the model into assistant and agent tasks that require step-by-step task breakdown before action.
  • Use standard Transformers code to generate responses with a preliminary thought block.
  • Test the model on math and coding tasks requiring detailed logical inference.

Example

Running response generation with logical inference in Python using transformers:

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

name = "kaivoss/LLaDA2.2-mini-Thinking"
tok = AutoTokenizer.from_pretrained(name, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(name, trust_remote_code=True, dtype=torch.bfloat16, device_map="cuda")

messages = [{"role": "user", "content": "Solve 5x^2 - 6x + 4 = 0."}]
prompt = tok.apply_chat_template(messages, add_generation_prompt=True, tokenize=False)
ids = tok(prompt, add_special_tokens=False, return_tensors="pt").input_ids.to("cuda")
out = model.generate(ids, gen_length=1536, block_length=32, steps=32, threshold=0.5, eos_early_stop=True)
text = "\n" + tok.decode(out[0], skip_special_tokens=True)
thinking, answer = text.split("", 1)
print(answer.strip())
Read moreDetails
Base: inclusionAI/LLaDA2.2-mini
Method: LoRA r32 on attention, dense MLPs, and shared experts with subsequent weight merging
VRAM requirements: about 33 GB for weights, an 80 GB VRAM GPU is sufficient
Limitations: evaluated on only 8 examples, long reasoning sequences may hit token length limits
Base model license: Apache-2.0
Discuss Telegram 0 25

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 30 GB · ready files31 GB
  • FP8 / INT8recommendedweights 15 GB16 GB
  • 4-bitweights 8.5 GB9.4 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 · BF16 — from RTX 5090

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.

    # kaivoss/LLaDA2.2-mini-Thinking — A30, FP8 / INT8
    # 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${VERSION_ID//./}/x86_64/cuda-keyring_1.1-1_all.deb" && \
      sudo dpkg -i cuda-keyring_1.1-1_all.deb && rm -f cuda-keyring_1.1-1_all.deb && \
      sudo apt-get update && \
      sudo DEBIAN_FRONTEND=noninteractive apt-get install -y nvidia-open || \
      { echo "Could not install the NVIDIA driver: see the error above. To keep the current one, run with KEEP_DRIVER=1."; exit 1; }
      echo "NVIDIA driver updated. Reboot the machine (sudo reboot) and run this same script again."
      exit 0
    fi
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv venv ~/vllm --python 3.12 && source ~/vllm/bin/activate
    uv pip install vllm
    vllm serve kaivoss/LLaDA2.2-mini-Thinking --max-model-len 8192 --quantization fp8
    Add more LoRAs to LLaDA2.2-mini-Thinking
  3. 3. Use it

    The server listens on port 8000 and speaks the OpenAI API: http://localhost:8000/v1 on the machine. From your computer: ssh -L 8000:localhost:8000 user@machine-IP.

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 LLaDA2.2-mini (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 6All on Hugging Face
Fine-tunes 2All on Hugging Face
LoRAs and adapters 1All on Hugging Face

About the model

kaivoss/LLaDA2.2-mini-Thinking
Author
kaivoss
Category
Language models
Released
Posted

On NeyroLearn

Hugging Face

250

Ask AI about this model

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

Comments