Skip to content
Language models

Laguna-XS.2-FP8

a compact MoE model for agentic coding

2 min readupdated

Laguna XS.2-FP8 is an FP8-quantized (8-bit floating point to save VRAM) version of Poolside's language model, built for software development and autonomous coding agents on local hardware. Under the hood, it uses a Mixture-of-Experts (MoE, where only part of the network activates per token) architecture: out of 33B total parameters, only 3B are active at a time, which speeds up generation. Across 30 of its 40 layers, it uses Sliding Window Attention (512-token sliding window) with per-head gating, while the KV cache is also compressed to FP8 to cut memory overhead. The model supports a context length of up to 131,072 tokens, reasons between tool calls, and fits developers looking for a capable local coding assistant under an open Apache 2.0 license.

Key features

  • Efficient MoE architecture: 33B total parameters with 3B active per token (256 experts plus 1 shared) and a 3:1 ratio of SWA to global attention layers.
  • Built-in reasoning mode: reasons before calling tools and between steps, with thinking toggleable on a per-request basis.
  • Ready for local deployment: despite its 33B size, it fits on a Mac with 36 GB RAM and runs out of the box in vLLM, Transformers, TRT-LLM, Ollama, and mlx-lm.
  • Permissive Apache 2.0 license: free for commercial and non-commercial use and modification.

What you can do

  • Run a local coding and terminal agent: hook it up in vLLM or Ollama with tool calling to execute bash commands, debug issues, and handle long multi-step tasks across a repo.
  • Tune reasoning to the task: keep preserved thinking enabled for complex architecture and refactoring, or toggle enable_thinking off for fast boilerplate generation.

Example

VLLM_USE_DEEP_GEMM=0 vllm serve \
    --model poolside/Laguna-XS.2-FP8 \
    --tool-call-parser poolside_v1 \
    --reasoning-parser poolside_v1 \
    --enable-auto-tool-choice \
    --default-chat-template-kwargs '{"enable_thinking": true}' \
    --kv-cache-dtype-skip-layers $(seq 0 39) \
    --moe_backend marlin
Read moreDetails
Benchmarks (for the BF16 version): SWE-bench Verified — 68.2%, SWE-bench Multilingual — 62.4%, SWE-bench Pro — 44.5%, Terminal-Bench 2.0 — 30.1%.
Hardware and caveats: on non-Hopper GPUs, the FP8 KV cache can produce garbled output — disable it in vLLM using the --kv-cache-dtype-skip-layers flag across all 40 layers.
Training: pre-training, post-training, and RL with the Muon optimizer; BF16, NVFP4, and INT4 variants are also available.
Discuss Telegram 0 38

Will it run?

Runs in 4-bit

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

  • F8_E4M3recommendedweights 33 GB · ready files35 GB
  • 4-bitweights 18 GB19 GB

Tip: go 8-bit.F8_E4M3 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: Won't fit any home GPU — renting is cheaper.

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 RTX A6000

    $0.43 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.

    # deepakku-redhat/Laguna-XS.2-FP8 — RTX A6000, F8_E4M3
    # 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 deepakku-redhat/Laguna-XS.2-FP8 --max-model-len 8192
    Add more LoRAs to Laguna-XS.2-FP8
  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 Laguna-XS.2-FP8 (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

deepakku-redhat/Laguna-XS.2-FP8
Author
deepakku-redhat
Category
Language models
Released
Posted

On NeyroLearn

Hugging Face

380

Ask AI about this model

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

Comments