AtomicChat/Laguna-S-2.1-GGUF

Base model: poolside/Laguna-S-2.1

Laguna S 2.1, self-quantized to GGUF by Atomic Chat. Built straight from Poolside's original weights with a per-tensor importance matrix, so this is not a repack of somebody else's files. Runs fully offline.

Highlights

117.6B parameters: the weights this repo quantizes.
Context length: 1,048,576 tokens (1M), as published by Poolside.
48 layers: Mixture-of-Experts, hybrid sliding-window (512) and global attention.
Full imatrix ladder: every quant is calibrated with an importance matrix, published here alongside the quants.
Mixed SWA and global attention layout: 48 layers in a 1:3 global-to-SWA ratio (12 global attention layers, 36 sliding-window layers, window 512), with softplus attention gating and per-layer-type rotary scales.
Native reasoning support: interleaved thinking between tool calls, with per-request control via enable_thinking.
Speculative decoding: a trained DFlash draft model is available for lower-latency serving.

[!NOTE] These GGUFs are self-quantized from the original weights, not a repack. The importance matrix keeps low-bit quants closer to the full-precision model.

[!IMPORTANT] Always pass --jinja so the Laguna S 2.1 chat template is applied. Without it the model can emit malformed turns.

Model Overview

Property	Value
Base model	`poolside/Laguna-S-2.1`
Parameters	117.6B
Layers	48
Experts	256 routed (top-10)
Sliding window	512 tokens
Context length	1,048,576 tokens (1M)
Vocabulary	100,352
Modalities	Text
Architecture	Mixture-of-Experts, 256 experts (top-10), hybrid sliding-window (512) and global attention, 48 attention heads over 8 KV heads, `LagunaForCausalLM`
This repo	GGUF quants (imatrix); the importance matrix is published here as `imatrix-coding.gguf`. Quants: `coding-IQ2_XS`, `coding-IQ2_M`, `coding-IQ3_M`, `coding-IQ4_XS`, `Q4_K_S`, `Q4_K_M`, `Q5_K_M`, `Q6_K`, `Q8_0`

Scores are Poolside's published results for the base poolside/Laguna-S-2.1, not our own measurements. Quantization preserves the large majority of this; Q4_K_M and up stay close to full precision.

Choosing a quant

Quant	Size	Notes
`coding-IQ2_XS`	34.4 GB	Very low memory.
`coding-IQ2_M`	38.4 GB	Very low memory, imatrix keeps it coherent.
`coding-IQ3_M`	51.5 GB	Beats Q3 at a similar size thanks to imatrix. Best low-RAM pick.
`coding-IQ4_XS`	62.7 GB	Excellent quality for size. Recommended low-bit.
`Q4_K_S`	66.9 GB	Compact 4-bit, fast.
`Q4_K_M`	71.2 GB	Recommended default. Best balance of size, speed and quality.
`Q5_K_M`	83.5 GB	Higher quality, low loss.
`Q6_K`	96.6 GB	Near lossless, noticeably lighter than Q8_0.
`Q8_0`	125.0 GB	Effectively lossless, reference quality.

[!TIP] Pick the largest file that fits your (V)RAM with room for context. Q4_K_M is the sweet spot for most setups; Q6_K or Q8_0 for maximum fidelity.

Get started

Run Laguna S 2.1 locally with:

Atomic Chat: the easiest path. Open the app, search AtomicChat/Laguna-S-2.1-GGUF, pick a quant, hit Use this model.
llama.cpp: llama-server -hf AtomicChat/Laguna-S-2.1-GGUF:Q4_K_M --jinja -c 8192
Ollama: ollama run hf.co/AtomicChat/Laguna-S-2.1-GGUF:Q4_K_M
LM Studio / Jan: search the repo id, download any quant.

Best practices

Parameter	Value
temperature	1.0
top_p	1.0
top_k	20
min_p	0.0

Poolside's recommended sampling configuration for poolside/Laguna-S-2.1.

Run in llama.cpp

git clone https://github.com/ggml-org/llama.cpp
cmake llama.cpp -B llama.cpp/build -DBUILD_SHARED_LIBS=OFF -DGGML_CUDA=ON
cmake --build llama.cpp/build --config Release -j --target llama-cli llama-server

./llama.cpp/build/bin/llama-server \
    -hf AtomicChat/Laguna-S-2.1-GGUF:Q4_K_M \
    --jinja -ngl 99 -c 8192 -fa on

How these were made

Download poolside/Laguna-S-2.1 (original weights).
Convert to f16 GGUF with llama.cpp.
Build an importance matrix over our calibration corpus, published here as imatrix-coding.gguf.
Quantize the ladder with --imatrix.

License

Original model by Poolside, released under the OpenMDW-1.1 license. Full terms: OpenMDW-1.1. Quantized by Atomic Chat.