sglang/docs/cookbook/autoregressive/Poolside/Laguna-XS-2.1.mdx

---
title: Laguna-XS-2.1
description: "Deploy poolside's Laguna-XS-2.1 — a 33B hybrid-SWA Mixture-of-Experts model (3B active) for agentic coding — with SGLang on NVIDIA H200, B300, and GB300 in BF16, FP8, NVFP4, and INT4."
---

## Deployment

<a id="install" />

<Accordion title="Install SGLang">

Laguna-XS-2.1 support is fully merged to SGLang `main` ([PR #29446](https://github.com/sgl-project/sglang/pull/29446): DFlash speculative decoding + shared-expert fix; [PR #29761](https://github.com/sgl-project/sglang/pull/29761): INT4 loader fix). Any build at or past their merge covers every cell below.

The model ships custom config code on the Hub, so `--trust-remote-code` is required (included in the launch commands).

<Tabs>

<Tab title="Python (pip / uv)">

```bash Command
pip install -U uv
uv venv --python 3.12 && source .venv/bin/activate

git clone https://github.com/sgl-project/sglang.git
cd sglang
uv pip install --prerelease=allow -e python
```

Then run the **Python** output of the command panel below in that environment.

</Tab>

<Tab title="Docker">

```bash Command
docker pull lmsysorg/sglang:latest
```

For how to launch the image, see [Install → Method 3: Using Docker](../../../docs/get-started/install#method-3-using-docker). Substitute the inner `sglang serve ...` with what the command generator below produces.

</Tab>

</Tabs>

</Accordion>

Pick your hardware + quantization + strategy to generate the launch command. The two serving strategies cover the common operating points:

- **Low-latency** — DFlash speculative decoding with a matched draft model. Pick for chat and interactive agents.
- **High-throughput** — plain serving. Best for batch workloads, where speculation's draft + rejection overhead costs more than it saves.

On the 8-GPU HGX platforms (H200 / B300), BF16 and NVFP4 run plain `--tp 8`; FP8 and INT4 run `--tp 8 --ep-size 8` because their quantization scales cannot shard the MoE 8-way (see [Configuration Tips](#2-configuration-tips)). The 4-GPU GB300 node runs plain `--tp 4` throughout.

import { Deployment } from "/src/snippets/_deployment.jsx";
import { config }     from "/src/snippets/configs/poolside/laguna-xs21.jsx";
import { benchmarks } from "/src/snippets/configs/poolside/laguna-xs21-benchmarks.jsx";

<Deployment config={config} benchmarks={benchmarks} />

## Playground

The Playground is where you experiment with **SGLang features beyond the verified matrix**. The Deploy panel above only emits combinations that have been signed off; the Playground lets you turn on additional knobs (TP degree, parsers) on top of whichever cell the Deploy panel is currently showing.

import { Playground } from "/src/snippets/_playground.jsx";

<Playground config={config} />

## 1. Model Introduction

[Laguna-XS-2.1](https://huggingface.co/poolside/Laguna-XS-2.1) is an open-weight **33B-parameter** hybrid sliding-window-attention MoE model (**~3B active per token**) from [poolside](https://poolside.ai), built for agentic coding and long-horizon software engineering — the extra-small sibling of [Laguna-M.1](./Laguna-M.1).

**Key Features:**

- **Sparse MoE**: 40 layers, 256 routed experts, top-8 routing.
- **Hybrid attention**: 30 sliding-window layers (window 512) interleaved with 10 full-attention layers; 48 Q / 8 KV heads.
- **Long context**: 262,144 tokens (RoPE + YaRN on the full-attention layers).
- **DFlash drafts**: matched draft models (5-layer, ~0.9 GB) ship per quantization for low-latency serving.
- **Hybrid reasoning**: `<think>…</think>` toggled per request via `chat_template_kwargs={"enable_thinking": …}`.

**Available quantizations:**

<table style={{width: "100%", borderCollapse: "collapse", tableLayout: "fixed"}}>
  <colgroup>
    <col style={{width: "14%"}} />
    <col style={{width: "43%"}} />
    <col style={{width: "43%"}} />
  </colgroup>
  <thead>
    <tr style={{borderBottom: "2px solid #d55816"}}>
      <th style={{textAlign: "left", padding: "10px 12px", fontWeight: 700}}>Precision</th>
      <th style={{textAlign: "left", padding: "10px 12px", fontWeight: 700}}>Target model</th>
      <th style={{textAlign: "left", padding: "10px 12px", fontWeight: 700}}>Draft model</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td style={{padding: "9px 12px", fontWeight: 500}}><strong>BF16</strong></td>
      <td style={{padding: "9px 12px"}}>[`poolside/Laguna-XS-2.1`](https://huggingface.co/poolside/Laguna-XS-2.1)</td>
      <td style={{padding: "9px 12px"}}>[`poolside/Laguna-XS-2.1-DFlash`](https://huggingface.co/poolside/Laguna-XS-2.1-DFlash)</td>
    </tr>
    <tr>
      <td style={{padding: "9px 12px", fontWeight: 500}}><strong>FP8</strong></td>
      <td style={{padding: "9px 12px"}}>[`poolside/Laguna-XS-2.1-FP8`](https://huggingface.co/poolside/Laguna-XS-2.1-FP8)</td>
      <td style={{padding: "9px 12px"}}>[`poolside/Laguna-XS-2.1-DFlash-FP8`](https://huggingface.co/poolside/Laguna-XS-2.1-DFlash-FP8)</td>
    </tr>
    <tr>
      <td style={{padding: "9px 12px", fontWeight: 500}}><strong>NVFP4</strong></td>
      <td style={{padding: "9px 12px"}}>[`poolside/Laguna-XS-2.1-NVFP4`](https://huggingface.co/poolside/Laguna-XS-2.1-NVFP4)</td>
      <td style={{padding: "9px 12px"}}>[`poolside/Laguna-XS-2.1-DFlash-NVFP4`](https://huggingface.co/poolside/Laguna-XS-2.1-DFlash-NVFP4)</td>
    </tr>
    <tr>
      <td style={{padding: "9px 12px", fontWeight: 500}}><strong>INT4</strong></td>
      <td style={{padding: "9px 12px"}}>[`poolside/Laguna-XS-2.1-INT4`](https://huggingface.co/poolside/Laguna-XS-2.1-INT4)</td>
      <td style={{padding: "9px 12px"}}>[`poolside/Laguna-XS-2.1-DFlash-INT4`](https://huggingface.co/poolside/Laguna-XS-2.1-DFlash-INT4)</td>
    </tr>
  </tbody>
</table>

The drafts themselves are small bf16 models, each *calibrated against its quantized target* — always pair a target with its matched draft (mixing precisions degrades accept-length).

**License:** Apache 2.0

**Resources:** [Hugging Face](https://huggingface.co/poolside/Laguna-XS-2.1) · [Release blog post](https://poolside.ai/blog/laguna-a-deeper-dive) · [API platform](https://platform.poolside.ai).

## 2. Configuration Tips

**Attention backend**

Leave `--attention-backend` unset for High-throughput cells — auto-select is correct (`fa3` on Hopper, `trtllm_mha` on Blackwell). With DFlash active, auto-select instead falls back to `flashinfer`, which breaks this hybrid-SWA model at `tp ≥ 4` on Blackwell (greedy GSM8K 76% → 28%), so the Low-latency commands pin the target backend explicitly. Leave `--speculative-draft-attention-backend` unset. Never use `triton` attention with Laguna (GSM8K 13%).

**Quantized checkpoints cap plain TP at 4**

`moe_intermediate_size=512` with FP8 block `[128,128]` / INT4 `group_size=128` scales cannot shard 8-way (512/8 = 64 < 128 granularity): FP8 fails at weight creation, INT4 crashes in the Marlin kernel, on any hardware. The generated 8-GPU FP8/INT4 commands therefore use `--tp 8 --ep-size 8` — expert parallelism keeps whole experts per rank, using all 8 GPUs on one instance. FP8 additionally needs `SGLANG_SHARED_EXPERT_TP1=1` (its shared expert is also block-quantized; INT4's stays bf16). Alternatives: plain `--tp 4`, or `--tp 4 --dp-size 2`. Accuracy is parallelism-independent within eval noise (verified tp1 ≡ tp4 on GB300 and tp4 ≡ tp8+ep8 on H200).

**DFlash memory**

Low-latency cells carry `--mem-fraction-static 0.7`: the default fraction OOMs in the draft vocab all-gather at `tp 4` on GB300. Dense cells use the default heuristic.

**INT4 is mixed-precision**

The INT4 checkpoint quantizes MoE layers in mixed 4-bit / 8-bit config groups. Builds older than [PR #29761](https://github.com/sgl-project/sglang/pull/29761) crash at load with `KeyError: 'Linear'`.

**Chat template**

On transformers ≥ 5.10 the standalone `chat_template.jinja` auto-loads — no flag needed (the server logs `Auto-detected template features: reasoning_parser=poolside_v1, ...`). On older transformers (≤ ~5.8) the `{% include %}` stub in `tokenizer_config.json` cannot resolve and the server silently falls back to a generic template — pass `--chat-template <model-dir>/chat_template.jinja` explicitly there.

**Thinking**

Off by default; opt in per request with `extra_body={"chat_template_kwargs": {"enable_thinking": True}}`. The template gates on `enable_thinking` — the generic `thinking` key is ignored.

**Served model id**

The server registers the model under whatever you pass to `--model-path`; a client's `model` field must match it (`poolside/Laguna-XS-2.1`, or the `-FP8` / `-NVFP4` / `-INT4` id).

## 3. Advanced Usage

### 3.1 DFlash Speculative Decoding

DFlash is a block-wise speculative decoder: the 5-layer draft proposes a block of tokens and the target verifies the whole block in one forward pass, so only target-approved tokens are emitted — output quality is the target's by construction (GSM8K matches dense within noise on every quantization). The speedup lever is **accept-length**, the number of draft tokens surviving verification per target step:

- Measured ~6 tokens/step at `tp 1`, ~4 at `tp 4` (greedy GSM8K, matched-precision pairs; ~3 under mixed reasoning-heavy traffic; FP8 reached 6.75 at `tp 8 + ep 8` on H200) — versus 1 token/step dense.
- Best for interactive / few-stream serving. Under batch-saturated load prefer High-throughput: once the GPU is compute-bound, draft + rejected-token overhead costs aggregate throughput.
- The generated commands always pair the draft calibrated for the selected target precision.

### 3.2 Reasoning

Launch with `--reasoning-parser poolside_v1` (baked into every generated command). Reasoning is opt-in via `enable_thinking=True`; the `<think>` trace lands in `message.reasoning_content`, separate from the final answer in `message.content`.

<Accordion title="Reasoning Example (Python)">

```python Example
from openai import OpenAI

client = OpenAI(base_url="http://localhost:30000/v1", api_key="EMPTY")

response = client.chat.completions.create(
    model="poolside/Laguna-XS-2.1",
    messages=[{"role": "user", "content": "What is 15% of 240? Explain briefly."}],
    max_tokens=2048,
    extra_body={"chat_template_kwargs": {"enable_thinking": True}},
)

message = response.choices[0].message
print("=============== Reasoning ===============")
print(message.reasoning_content)
print("=============== Answer ==================")
print(message.content)
```

</Accordion>

<Note>
XS-2.1 is an extra-small model — give it generous `max_tokens` when thinking is enabled
(hard problems regularly reason for thousands of tokens), and keep thinking off for
short-form tasks.
</Note>

### 3.3 Tool Calling

Launch with `--tool-call-parser poolside_v1` (baked into every generated command). The parser converts Laguna's `<tool_call>` output into the standard OpenAI `tool_calls` structure. Tool calling works with reasoning off (the default).

<Accordion title="Tool Calling Example (Python)">

```python Example
from openai import OpenAI

client = OpenAI(base_url="http://localhost:30000/v1", api_key="EMPTY")

tools = [
    {
        "type": "function",
        "function": {
            "name": "get_weather",
            "description": "Get the current weather for a location",
            "parameters": {
                "type": "object",
                "properties": {
                    "location": {"type": "string", "description": "The city name"},
                    "unit": {"type": "string", "enum": ["celsius", "fahrenheit"]},
                },
                "required": ["location"],
            },
        },
    }
]

response = client.chat.completions.create(
    model="poolside/Laguna-XS-2.1",
    messages=[{"role": "user", "content": "What's the weather in Beijing?"}],
    tools=tools,
)

message = response.choices[0].message
if message.tool_calls:
    for call in message.tool_calls:
        print(f"Tool: {call.function.name}")
        print(f"Args: {call.function.arguments}")
```

</Accordion>