diff --git a/docs_new/cookbook/autoregressive/Poolside/Laguna-M.1.mdx b/docs_new/cookbook/autoregressive/Poolside/Laguna-M.1.mdx index ea643df4a..32976dcbe 100644 --- a/docs_new/cookbook/autoregressive/Poolside/Laguna-M.1.mdx +++ b/docs_new/cookbook/autoregressive/Poolside/Laguna-M.1.mdx @@ -1,7 +1,6 @@ --- title: Laguna-M.1 description: "Deploy poolside's Laguna-M.1 — a 225B-parameter Mixture-of-Experts model (23B active) for agentic coding — with SGLang on NVIDIA H200, B200, B300, GB200, and GB300, across BF16, FP8, and NVFP4." -tag: NEW --- ## Deployment diff --git a/docs_new/cookbook/autoregressive/Poolside/Laguna-S-2.1.mdx b/docs_new/cookbook/autoregressive/Poolside/Laguna-S-2.1.mdx new file mode 100644 index 000000000..1d4f13b36 --- /dev/null +++ b/docs_new/cookbook/autoregressive/Poolside/Laguna-S-2.1.mdx @@ -0,0 +1,244 @@ +--- +title: Laguna-S-2.1 +description: "Deploy poolside's Laguna-S-2.1 — a 118B hybrid-SWA Mixture-of-Experts model (8B active) for agentic coding — with SGLang on NVIDIA H200, B300, and GB300 in BF16, FP8, NVFP4, and INT4." +tag: NEW +--- + +## Deployment + + + + + +Laguna-S-2.1 uses the same `laguna` model architecture as [Laguna-XS-2.1](./Laguna-XS-2.1), which is fully supported in SGLang `main`. The model ships custom config code on the Hub, so `--trust-remote-code` is required (included in the launch commands). + + + + + +```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 -e python +``` + +Then run the **Python** output of the command panel below in that environment. + + + + + +```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. + + + + + + + +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) all quantizations run `--tp 8`. The 4-GPU GB300 node runs `--tp 4` throughout. NVFP4 is Blackwell-only (B300 / GB300 only). + +import { Deployment } from "/src/snippets/_deployment.jsx"; +import { config } from "/src/snippets/configs/poolside/laguna-s21.jsx"; +import { benchmarks } from "/src/snippets/configs/poolside/laguna-s21-benchmarks.jsx"; + + + +## 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"; + + + +## 1. Model Introduction + +[Laguna-S-2.1](https://huggingface.co/poolside/Laguna-S-2.1) is an open-weight **118B-parameter** hybrid sliding-window-attention MoE model (**~8B active per token**) from [poolside](https://poolside.ai), built for agentic coding and long-horizon software engineering. It sits between [Laguna XS 2.1](./Laguna-XS-2.1) (33B/3B active) and Laguna M.1 (222B/23B active) in the Laguna family. + +**Key Features:** + +- **Sparse MoE**: 48 layers, 256 routed experts, top-10 routing, plus 1 shared expert. +- **Hybrid attention**: 36 sliding-window layers (window 512) interleaved with 12 full-attention layers (1:3 global-to-SWA ratio); 8 KV heads, head dim 128; per-head sigmoid output gating with per-layer-type rotary scales. +- **Long context**: 1,048,576 tokens. +- **DFlash drafts**: matched draft models ship per quantization for low-latency serving. +- **Hybrid reasoning**: `…` toggled per request via `chat_template_kwargs={"enable_thinking": …}`. + +**Available quantizations:** + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +
PrecisionTarget modelDraft model
BF16[`poolside/Laguna-S-2.1`](https://huggingface.co/poolside/Laguna-S-2.1)[`poolside/Laguna-S-2.1-DFlash`](https://huggingface.co/poolside/Laguna-S-2.1-DFlash)
FP8[`poolside/Laguna-S-2.1-FP8`](https://huggingface.co/poolside/Laguna-S-2.1-FP8)[`poolside/Laguna-S-2.1-DFlash-FP8`](https://huggingface.co/poolside/Laguna-S-2.1-DFlash-FP8)
NVFP4[`poolside/Laguna-S-2.1-NVFP4`](https://huggingface.co/poolside/Laguna-S-2.1-NVFP4)[`poolside/Laguna-S-2.1-DFlash-NVFP4`](https://huggingface.co/poolside/Laguna-S-2.1-DFlash-NVFP4)
INT4[`poolside/Laguna-S-2.1-INT4`](https://huggingface.co/poolside/Laguna-S-2.1-INT4)[`poolside/Laguna-S-2.1-DFlash-INT4`](https://huggingface.co/poolside/Laguna-S-2.1-DFlash-INT4)
+ +The drafts are small BF16 models, each *calibrated against its quantized target* — always pair a target with its matched draft (mixing precisions degrades accept-length). + +**License:** [OpenMDW-1.1](https://openmdw.ai/) + +**Resources:** [Hugging Face](https://huggingface.co/poolside/Laguna-S-2.1) · [Technical report](https://poolside.ai/assets/laguna/laguna-m1-xs2-technical-report.pdf) · [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 (reproduced on Laguna-XS-2.1, greedy GSM8K 76% → 28%), so the Low-latency commands pin the target backend explicitly. Leave `--speculative-draft-attention-backend` unset. Other attention backend choices have not been fully validated on Laguna; keep the default. + +**BF16 memory on H200** + +BF16 on H200 leaves less headroom for CUDA-graph capture and NCCL allocations than FP8/INT4. The High-throughput BF16 command carries `--mem-fraction-static 0.80`. FP8, INT4, and all B300/GB300 cells use the default heuristic. + +**FP8 shared expert** + +`SGLANG_SHARED_EXPERT_TP1=1` is required for FP8 cells on **all hardware** — confirmed on both H200 (TP=8) and GB300 (TP=4). The FP8 checkpoint block-quantizes the shared expert (128×128 scales), which cannot TP-shard cleanly at either TP degree on S-2.1. This env var replicates the shared expert instead of sharding it. INT4 keeps the shared expert in BF16 (no flag needed); BF16 is unquantized. Note: this differs from Laguna-XS-2.1 where TP=4 does not require the flag — the constraint is architecture-specific. + +**FP8 and NVFP4 DFlash drafts** + +Fixed upstream on 2026-07-21: all DFlash draft configs now use a flat top-level `rope_theta` (the `rope_parameters` block was removed). If a server crashes at draft-model load with `KeyError: 'rope_theta'`, you are serving a draft checkpoint cached before 2026-07-21 — re-download it (e.g. `hf download poolside/Laguna-S-2.1-DFlash-FP8`) to pick up the corrected config. + +**DFlash memory** + +Low-latency cells carry `--mem-fraction-static 0.7` (sufficient even for BF16 on H200). Dense cells use the default heuristic (except BF16 on H200 — see above). + +**BF16 reasoning length** + +BF16 reasons approximately 2× longer than FP8/INT4 on AIME25 (median 34.8 k vs 16.9 k tokens), consistently truncating at `max_tokens=64000`. FP8/INT4 truncate at ≈ 2%. For a valid BF16 AIME25 score, serve with `max_tokens ≥ 131072` (the model supports a 1 M context window). + +**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) pass `--chat-template /chat_template.jinja` explicitly. + +**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-S-2.1`, or the `-FP8` / `-NVFP4` / `-INT4` id). + +## 3. Advanced Usage + +### 3.1 DFlash Speculative Decoding + +DFlash is a block-wise speculative decoder: the draft proposes a block of tokens and the target verifies the whole block in one forward pass — output quality is the target's by construction. The speedup lever is **accept-length**, the number of draft tokens surviving verification per target step. + +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 `` trace lands in `message.reasoning_content`, separate from the final answer in `message.content`. + + + +```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-S-2.1", + messages=[{"role": "user", "content": "What is 15% of 240? Explain briefly."}], + max_tokens=4096, + extra_body={"chat_template_kwargs": {"enable_thinking": True}}, +) + +message = response.choices[0].message +print("=============== Reasoning ===============") +print(message.reasoning_content) +print("=============== Answer ==================") +print(message.content) +``` + + + + +Give generous `max_tokens` when thinking is enabled — hard problems regularly reason +for thousands of tokens. Keep thinking off for short-form tasks. + + +### 3.3 Tool Calling + +Launch with `--tool-call-parser poolside_v1` (baked into every generated command). The parser converts Laguna's `` output into the standard OpenAI `tool_calls` structure. Tool calling works with reasoning off (the default). + + + +```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-S-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}") +``` + + diff --git a/docs_new/cookbook/autoregressive/Poolside/Laguna-XS-2.1.mdx b/docs_new/cookbook/autoregressive/Poolside/Laguna-XS-2.1.mdx index 5423eeda2..7d0ff41fa 100644 --- a/docs_new/cookbook/autoregressive/Poolside/Laguna-XS-2.1.mdx +++ b/docs_new/cookbook/autoregressive/Poolside/Laguna-XS-2.1.mdx @@ -1,7 +1,6 @@ --- 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." -tag: NEW --- ## Deployment diff --git a/docs_new/cookbook/autoregressive/intro.mdx b/docs_new/cookbook/autoregressive/intro.mdx index 5dac02921..3731e8e3f 100644 --- a/docs_new/cookbook/autoregressive/intro.mdx +++ b/docs_new/cookbook/autoregressive/intro.mdx @@ -160,7 +160,7 @@ metatags: diff --git a/docs_new/docs.json b/docs_new/docs.json index 1c2d81bcf..c01f7bb5e 100644 --- a/docs_new/docs.json +++ b/docs_new/docs.json @@ -1172,6 +1172,7 @@ { "group": "Poolside", "pages": [ + "cookbook/autoregressive/Poolside/Laguna-S-2.1", "cookbook/autoregressive/Poolside/Laguna-XS-2.1", "cookbook/autoregressive/Poolside/Laguna-M.1", "cookbook/autoregressive/Poolside/Laguna-XS.2" diff --git a/docs_new/src/snippets/configs/poolside/laguna-s21-benchmarks.jsx b/docs_new/src/snippets/configs/poolside/laguna-s21-benchmarks.jsx new file mode 100644 index 000000000..73fdc3589 --- /dev/null +++ b/docs_new/src/snippets/configs/poolside/laguna-s21-benchmarks.jsx @@ -0,0 +1,218 @@ +export const benchmarks = [ + + // ── H200 (8×H200, tp 8, sglang 0.5.15.post1, random ISL=8192/OSL=1024) ── + // tokens_per_sec_per_gpu = (input+output) tok/s/GPU = output_tok_s / 8 * (8192+1024)/1024 + // TTFT/TPOT are mean values from bench_serving. + + { + match: { hw: "h200", variant: "default", quant: "bf16", strategy: "high-throughput", nodes: "single" }, + verified: true, + sglang_version: "0.5.15.post1", + speed: [ + { workload: { dataset: "random", isl: 8192, osl: 1024, max_concurrency: 1024 }, + ttft_ms: 23742, tpot_ms: 256.8, tokens_per_sec_per_gpu: 5264 }, + ], + // BF16 reasons ~2× longer than FP8/INT4 (median 34.8k vs 16.9k tokens); + // truncation at max_tokens=64000 invalidates the result. Needs max_tokens ≥ 131072. + accuracy: { gsm8k_pct: 93.18, aime25_pct: null }, + }, + { + match: { hw: "h200", variant: "default", quant: "bf16", strategy: "low-latency", nodes: "single" }, + verified: true, + sglang_version: "0.5.15.post1", + speed: [ + { workload: { dataset: "random", isl: 8192, osl: 1024, max_concurrency: 1 }, + ttft_ms: 102.3, tpot_ms: 3.48, tokens_per_sec_per_gpu: 305 }, + { workload: { dataset: "random", isl: 8192, osl: 1024, max_concurrency: 16 }, + ttft_ms: 97.7, tpot_ms: 6.45, tokens_per_sec_per_gpu: 2094 }, + ], + accuracy: { gsm8k_pct: 93.33, aime25_pct: null }, + }, + { + match: { hw: "h200", variant: "default", quant: "fp8", strategy: "high-throughput", nodes: "single" }, + verified: true, + sglang_version: "0.5.15.post1", + speed: [ + { workload: { dataset: "random", isl: 8192, osl: 1024, max_concurrency: 1024 }, + ttft_ms: 16199, tpot_ms: 293.2, tokens_per_sec_per_gpu: 5175 }, + ], + accuracy: { gsm8k_pct: 94.24, aime25_pct: 0.654 }, + }, + { + match: { hw: "h200", variant: "default", quant: "fp8", strategy: "low-latency", nodes: "single" }, + verified: true, + sglang_version: "0.5.15.post1", + speed: [ + { workload: { dataset: "random", isl: 8192, osl: 1024, max_concurrency: 1 }, + ttft_ms: 97.7, tpot_ms: 4.43, tokens_per_sec_per_gpu: 242 }, + { workload: { dataset: "random", isl: 8192, osl: 1024, max_concurrency: 16 }, + ttft_ms: 99.4, tpot_ms: 7.16, tokens_per_sec_per_gpu: 1956 }, + ], + accuracy: { gsm8k_pct: 94.47, aime25_pct: null }, + }, + { + match: { hw: "h200", variant: "default", quant: "int4", strategy: "high-throughput", nodes: "single" }, + verified: true, + sglang_version: "0.5.15.post1", + speed: [ + { workload: { dataset: "random", isl: 8192, osl: 1024, max_concurrency: 1024 }, + ttft_ms: 19886, tpot_ms: 318.5, tokens_per_sec_per_gpu: 5055 }, + ], + accuracy: { gsm8k_pct: 95.00, aime25_pct: 0.690 }, + }, + { + match: { hw: "h200", variant: "default", quant: "int4", strategy: "low-latency", nodes: "single" }, + verified: true, + sglang_version: "0.5.15.post1", + speed: [ + { workload: { dataset: "random", isl: 8192, osl: 1024, max_concurrency: 1 }, + ttft_ms: 113.1, tpot_ms: 3.83, tokens_per_sec_per_gpu: 276 }, + { workload: { dataset: "random", isl: 8192, osl: 1024, max_concurrency: 16 }, + ttft_ms: 94.1, tpot_ms: 6.40, tokens_per_sec_per_gpu: 2125 }, + ], + accuracy: { gsm8k_pct: 94.24, aime25_pct: null }, + }, + + // ── B300 (8×B300, tp 8, sglang 0.5.15.post1) ── + // AIME25 (high-throughput cells): sgl-eval run aime25, 30 problems × 16 repeats, + // temp 1.0, top_p 0.95, max_tokens 64000 (BF16: 131072), 128 threads, thinking ON + // via enable_thinking-patched sgl-eval. 2026-07-21. + + { + match: { hw: "b300", variant: "default", quant: "bf16", strategy: "high-throughput", nodes: "single" }, + verified: true, + sglang_version: "0.5.15.post1", + accuracy: { gsm8k_pct: 93.71, aime25_pct: null }, + }, + { + match: { hw: "b300", variant: "default", quant: "bf16", strategy: "low-latency", nodes: "single" }, + verified: true, + sglang_version: "0.5.15.post1", + accuracy: { gsm8k_pct: 93.63, aime25_pct: null }, + }, + { + match: { hw: "b300", variant: "default", quant: "fp8", strategy: "high-throughput", nodes: "single" }, + verified: true, + sglang_version: "0.5.15.post1", + accuracy: { gsm8k_pct: 94.39, aime25_pct: 66.88 }, + }, + { + match: { hw: "b300", variant: "default", quant: "fp8", strategy: "low-latency", nodes: "single" }, + verified: true, + sglang_version: "0.5.15.post1", + accuracy: { gsm8k_pct: 94.69, aime25_pct: null }, + }, + { + match: { hw: "b300", variant: "default", quant: "nvfp4", strategy: "high-throughput", nodes: "single" }, + verified: true, + sglang_version: "0.5.15.post1", + accuracy: { gsm8k_pct: 94.54, aime25_pct: 66.46 }, + }, + { + match: { hw: "b300", variant: "default", quant: "nvfp4", strategy: "low-latency", nodes: "single" }, + verified: true, + sglang_version: "0.5.15.post1", + accuracy: { gsm8k_pct: 95.30, aime25_pct: null }, + }, + { + match: { hw: "b300", variant: "default", quant: "int4", strategy: "high-throughput", nodes: "single" }, + verified: true, + sglang_version: "0.5.15.post1", + accuracy: { gsm8k_pct: 94.62, aime25_pct: 68.33 }, + }, + { + match: { hw: "b300", variant: "default", quant: "int4", strategy: "low-latency", nodes: "single" }, + verified: true, + sglang_version: "0.5.15.post1", + accuracy: { gsm8k_pct: 94.69, aime25_pct: null }, + }, + + // ── GB300 (4×GB300, tp 4, sglang 0.5.15.post1, random ISL=8192/OSL=1024) ── + // tokens_per_sec_per_gpu = output_tok_s / 4 * (8192+1024)/1024 + // TTFT/TPOT are median values from bench_serving. + + { + match: { hw: "gb300", variant: "default", quant: "bf16", strategy: "high-throughput", nodes: "single" }, + verified: true, + sglang_version: "0.5.15.post1", + accuracy: { gsm8k_pct: 93.33, aime25_pct: null }, + }, + { + match: { hw: "gb300", variant: "default", quant: "bf16", strategy: "low-latency", nodes: "single" }, + verified: true, + sglang_version: "0.5.15.post1", + speed: [ + { workload: { dataset: "random", isl: 8192, osl: 1024, max_concurrency: 1 }, + ttft_ms: 114, tpot_ms: 4.2, tokens_per_sec_per_gpu: 562 }, + { workload: { dataset: "random", isl: 8192, osl: 1024, max_concurrency: 16 }, + ttft_ms: 141, tpot_ms: 8.4, tokens_per_sec_per_gpu: 3413 }, + ], + accuracy: { gsm8k_pct: 93.33, aime25_pct: null }, + }, + { + match: { hw: "gb300", variant: "default", quant: "fp8", strategy: "high-throughput", nodes: "single" }, + verified: true, + sglang_version: "0.5.15.post1", + speed: [ + { workload: { dataset: "random", isl: 8192, osl: 1024, max_concurrency: 1024 }, + ttft_ms: 10062, tpot_ms: 216, tokens_per_sec_per_gpu: 9536 }, + ], + accuracy: { gsm8k_pct: 94.31, aime25_pct: null }, + }, + { + match: { hw: "gb300", variant: "default", quant: "fp8", strategy: "low-latency", nodes: "single" }, + verified: true, + sglang_version: "0.5.15.post1", + speed: [ + { workload: { dataset: "random", isl: 8192, osl: 1024, max_concurrency: 1 }, + ttft_ms: 109, tpot_ms: 6.1, tokens_per_sec_per_gpu: 369 }, + { workload: { dataset: "random", isl: 8192, osl: 1024, max_concurrency: 16 }, + ttft_ms: 135, tpot_ms: 10.7, tokens_per_sec_per_gpu: 2578 }, + ], + accuracy: { gsm8k_pct: 94.54, aime25_pct: null }, + }, + { + match: { hw: "gb300", variant: "default", quant: "nvfp4", strategy: "high-throughput", nodes: "single" }, + verified: true, + sglang_version: "0.5.15.post1", + speed: [ + { workload: { dataset: "random", isl: 8192, osl: 1024, max_concurrency: 1024 }, + ttft_ms: 7186, tpot_ms: 211, tokens_per_sec_per_gpu: 9658 }, + ], + accuracy: { gsm8k_pct: 94.47, aime25_pct: null }, + }, + { + match: { hw: "gb300", variant: "default", quant: "nvfp4", strategy: "low-latency", nodes: "single" }, + verified: true, + sglang_version: "0.5.15.post1", + speed: [ + { workload: { dataset: "random", isl: 8192, osl: 1024, max_concurrency: 1 }, + ttft_ms: 115, tpot_ms: 11.2, tokens_per_sec_per_gpu: 204 }, + { workload: { dataset: "random", isl: 8192, osl: 1024, max_concurrency: 16 }, + ttft_ms: 125, tpot_ms: 18.8, tokens_per_sec_per_gpu: 1549 }, + ], + accuracy: { gsm8k_pct: 94.77, aime25_pct: null }, + }, + { + match: { hw: "gb300", variant: "default", quant: "int4", strategy: "high-throughput", nodes: "single" }, + verified: true, + sglang_version: "0.5.15.post1", + speed: [ + { workload: { dataset: "random", isl: 8192, osl: 1024, max_concurrency: 1024 }, + ttft_ms: 10428, tpot_ms: 216, tokens_per_sec_per_gpu: 9184 }, + ], + accuracy: { gsm8k_pct: 94.69, aime25_pct: null }, + }, + { + match: { hw: "gb300", variant: "default", quant: "int4", strategy: "low-latency", nodes: "single" }, + verified: true, + sglang_version: "0.5.15.post1", + speed: [ + { workload: { dataset: "random", isl: 8192, osl: 1024, max_concurrency: 1 }, + ttft_ms: 134, tpot_ms: 5.4, tokens_per_sec_per_gpu: 410 }, + { workload: { dataset: "random", isl: 8192, osl: 1024, max_concurrency: 16 }, + ttft_ms: 127, tpot_ms: 10.1, tokens_per_sec_per_gpu: 2880 }, + ], + accuracy: { gsm8k_pct: 95.00, aime25_pct: null }, + }, +]; diff --git a/docs_new/src/snippets/configs/poolside/laguna-s21.jsx b/docs_new/src/snippets/configs/poolside/laguna-s21.jsx new file mode 100644 index 000000000..cdf9ac360 --- /dev/null +++ b/docs_new/src/snippets/configs/poolside/laguna-s21.jsx @@ -0,0 +1,506 @@ +// Laguna-S-2.1 (poolside) — 118B MoE (8B active), 1M context, laguna arch (SGLang main). +// --trust-remote-code required (custom config code on Hub). +// +// Attention backend: leave unset for High-Throughput (auto-selects fa3/trtllm_mha). +// With DFlash active, auto falls back to flashinfer which breaks hybrid-SWA at tp≥4 +// on Blackwell — Low-Latency cells pin the target backend explicitly. +// Never use --attention-backend triton on Laguna (broken SWA handling). +// +// BF16 on H200 HT: --mem-fraction-static 0.80 required — BF16 leaves less headroom +// for CUDA-graph capture and NCCL allocs than FP8/INT4 on 141 GB/GPU. +// B300/GB300 (288 GB/GPU) unaffected. +// +// SGLANG_SHARED_EXPERT_TP1=1 (FP8 cells, all hardware): FP8 block-quantizes the shared +// expert; required at both TP=4 (GB300) and TP=8 (H200/B300). INT4 shared expert stays +// bf16 — no flag needed. Differs from Laguna-XS-2.1 where TP=4 works without this flag. +// +// NVFP4 is Blackwell-only → no h200×nvfp4 cells. +// DFlash cells carry --mem-fraction-static 0.7. + +export const config = { + modelName: "Laguna-S-2.1", + + supportedHardware: ["h200", "b300", "gb300"], + + variants: [ + { id: "default", label: "Default" }, + ], + + quantizations: [ + { id: "bf16", label: "BF16" }, + { id: "fp8", label: "FP8" }, + { id: "nvfp4", label: "NVFP4" }, + { id: "int4", label: "INT4" }, + ], + + strategies: [ + { id: "low-latency", label: "Low-latency" }, + { id: "high-throughput", label: "High-throughput" }, + ], + + nodesOptions: [ + { id: "single", label: "Single Node" }, + ], + + modelNames: { + "default|bf16": "poolside/Laguna-S-2.1", + "default|fp8": "poolside/Laguna-S-2.1-FP8", + "default|nvfp4": "poolside/Laguna-S-2.1-NVFP4", + "default|int4": "poolside/Laguna-S-2.1-INT4", + }, + + placeholders: { + HOST_IP: { target: "command", label: "Bind host", default: "0.0.0.0" }, + PORT: { target: "command", label: "Bind port", default: "30000" }, + HF_TOKEN: { target: "command", label: "HF token (Docker)", default: "" }, + CURL_HOST: { target: "curl", label: "Server host", default: "localhost" }, + CURL_PORT: { target: "curl", label: "Server port", default: "30000" }, + }, + + curl: `curl http://{{CURL_HOST}}:{{CURL_PORT}}/v1/chat/completions \\ +-H 'Content-Type: application/json' \\ +-d '{ "model": "{{MODEL_NAME}}", "messages": [{"role":"user","content":"Hello"}] }'`, + + benchmarkCommands: { + speed: +`python3 -m sglang.bench_serving \\ + --backend sglang \\ + --host {{CURL_HOST}} --port {{CURL_PORT}} \\ + --model {{MODEL_NAME}} \\ + --dataset-name {{DATASET}} \\ + --random-input-len {{ISL}} --random-output-len {{OSL}} \\ + --num-prompts {{NUM_PROMPTS}} --max-concurrency {{MAX_CONCURRENCY}}`, + accuracy: { + gsm8k_pct: +`# pip install git+https://github.com/sgl-project/sgl-eval +sgl-eval run gsm8k \\ + --base-url http://{{CURL_HOST}}:{{CURL_PORT}}/v1 \\ + --num-threads 128`, + // Laguna's template gates on enable_thinking, not the generic 'thinking' key. + // Serve with a copy of the model's chat template whose enable_thinking default + // is flipped to true. For BF16: use --max-tokens 131072 (see Configuration Tips). + aime25_pct: +`# pip install git+https://github.com/sgl-project/sgl-eval +# Serve with an enable_thinking=true chat template (see Configuration Tips: Thinking). +# For BF16: use --max-tokens 131072 (see Configuration Tips: BF16 reasoning length). +sgl-eval run aime25 \\ + --base-url http://{{CURL_HOST}}:{{CURL_PORT}}/v1 \\ + --n-repeats 16 --max-tokens 64000 \\ + --temperature 1.0 --top-p 0.95 --thinking \\ + --num-threads 128`, + }, + numPromptsByConc: { 1: 8, 16: 32, 64: 128, 128: 256, 256: 512, 1024: 2048, 4096: 4096 }, + }, + + defaultAccuracy: { + default: { gsm8k_pct: null, aime25_pct: null }, + }, + + accuracyLabels: [ + ["gsm8k_pct", "GSM8K", "%"], + ["aime25_pct", "AIME25", "%"], + ], + + dockerImages: { + h200: "lmsysorg/sglang:latest", + b300: "lmsysorg/sglang:dev", + gb300: "lmsysorg/sglang:dev", + }, + + github: { + cookbookModel: "poolside/Laguna-S-2.1", + }, + + playgroundFeatures: { + attention: { + knobs: [ + { id: "tp", label: "TP", values: [null, 1, 2, 4, 8] }, + ], + }, + parsers: { + items: [ + { id: "reasoning", label: "Reasoning Parser", flag: "--reasoning-parser poolside_v1" }, + { id: "toolCall", label: "Tool Call Parser", flag: "--tool-call-parser poolside_v1" }, + ], + }, + }, + + cells: [ + + // ══════════════ B300 FP8 low-latency — default (cells[0]) ══════════════ + + { + match: { hw: "b300", variant: "default", quant: "fp8", strategy: "low-latency", nodes: "single" }, + verified: true, + env: ["SGLANG_SHARED_EXPERT_TP1=1"], + flags: [ + "--model-path {{MODEL_NAME}}", + "--trust-remote-code", + "--reasoning-parser poolside_v1", + "--tool-call-parser poolside_v1", + "--tp 8", + "--attention-backend trtllm_mha", + "--speculative-algorithm DFLASH", + "--speculative-draft-model-path poolside/Laguna-S-2.1-DFlash-FP8", + "--page-size 1", + "--mem-fraction-static 0.7", + "--host {{HOST_IP}}", + "--port {{PORT}}", + ], + }, + + // ══════════════ H200 (8-GPU HGX, tp 8) ══════════════ + + { + match: { hw: "h200", variant: "default", quant: "bf16", strategy: "high-throughput", nodes: "single" }, + verified: true, + env: [], + flags: [ + "--model-path {{MODEL_NAME}}", + "--trust-remote-code", + "--reasoning-parser poolside_v1", + "--tool-call-parser poolside_v1", + "--tp 8", + "--mem-fraction-static 0.80", + "--host {{HOST_IP}}", + "--port {{PORT}}", + ], + }, + { + match: { hw: "h200", variant: "default", quant: "bf16", strategy: "low-latency", nodes: "single" }, + verified: true, + env: [], + flags: [ + "--model-path {{MODEL_NAME}}", + "--trust-remote-code", + "--reasoning-parser poolside_v1", + "--tool-call-parser poolside_v1", + "--tp 8", + "--attention-backend fa3", + "--speculative-algorithm DFLASH", + "--speculative-draft-model-path poolside/Laguna-S-2.1-DFlash", + "--page-size 1", + "--mem-fraction-static 0.7", + "--host {{HOST_IP}}", + "--port {{PORT}}", + ], + }, + { + match: { hw: "h200", variant: "default", quant: "fp8", strategy: "high-throughput", nodes: "single" }, + verified: true, + env: ["SGLANG_SHARED_EXPERT_TP1=1"], + flags: [ + "--model-path {{MODEL_NAME}}", + "--trust-remote-code", + "--reasoning-parser poolside_v1", + "--tool-call-parser poolside_v1", + "--tp 8", + "--host {{HOST_IP}}", + "--port {{PORT}}", + ], + }, + { + match: { hw: "h200", variant: "default", quant: "fp8", strategy: "low-latency", nodes: "single" }, + verified: true, + env: ["SGLANG_SHARED_EXPERT_TP1=1"], + flags: [ + "--model-path {{MODEL_NAME}}", + "--trust-remote-code", + "--reasoning-parser poolside_v1", + "--tool-call-parser poolside_v1", + "--tp 8", + "--attention-backend fa3", + "--speculative-algorithm DFLASH", + "--speculative-draft-model-path poolside/Laguna-S-2.1-DFlash-FP8", + "--page-size 1", + "--mem-fraction-static 0.7", + "--host {{HOST_IP}}", + "--port {{PORT}}", + ], + }, + { + match: { hw: "h200", variant: "default", quant: "int4", strategy: "high-throughput", nodes: "single" }, + verified: true, + env: [], + flags: [ + "--model-path {{MODEL_NAME}}", + "--trust-remote-code", + "--reasoning-parser poolside_v1", + "--tool-call-parser poolside_v1", + "--tp 8", + "--host {{HOST_IP}}", + "--port {{PORT}}", + ], + }, + { + match: { hw: "h200", variant: "default", quant: "int4", strategy: "low-latency", nodes: "single" }, + verified: true, + env: [], + flags: [ + "--model-path {{MODEL_NAME}}", + "--trust-remote-code", + "--reasoning-parser poolside_v1", + "--tool-call-parser poolside_v1", + "--tp 8", + "--attention-backend fa3", + "--speculative-algorithm DFLASH", + "--speculative-draft-model-path poolside/Laguna-S-2.1-DFlash-INT4", + "--page-size 1", + "--mem-fraction-static 0.7", + "--host {{HOST_IP}}", + "--port {{PORT}}", + ], + }, + + // ══════════════ B300 (8-GPU HGX, tp 8) ══════════════ + + { + match: { hw: "b300", variant: "default", quant: "bf16", strategy: "high-throughput", nodes: "single" }, + verified: true, + env: [], + flags: [ + "--model-path {{MODEL_NAME}}", + "--trust-remote-code", + "--reasoning-parser poolside_v1", + "--tool-call-parser poolside_v1", + "--tp 8", + "--host {{HOST_IP}}", + "--port {{PORT}}", + ], + }, + { + match: { hw: "b300", variant: "default", quant: "bf16", strategy: "low-latency", nodes: "single" }, + verified: true, + env: [], + flags: [ + "--model-path {{MODEL_NAME}}", + "--trust-remote-code", + "--reasoning-parser poolside_v1", + "--tool-call-parser poolside_v1", + "--tp 8", + "--attention-backend trtllm_mha", + "--speculative-algorithm DFLASH", + "--speculative-draft-model-path poolside/Laguna-S-2.1-DFlash", + "--page-size 1", + "--mem-fraction-static 0.7", + "--host {{HOST_IP}}", + "--port {{PORT}}", + ], + }, + { + match: { hw: "b300", variant: "default", quant: "fp8", strategy: "high-throughput", nodes: "single" }, + verified: true, + env: ["SGLANG_SHARED_EXPERT_TP1=1"], + flags: [ + "--model-path {{MODEL_NAME}}", + "--trust-remote-code", + "--reasoning-parser poolside_v1", + "--tool-call-parser poolside_v1", + "--tp 8", + "--host {{HOST_IP}}", + "--port {{PORT}}", + ], + }, + { + match: { hw: "b300", variant: "default", quant: "nvfp4", strategy: "high-throughput", nodes: "single" }, + verified: true, + env: [], + flags: [ + "--model-path {{MODEL_NAME}}", + "--trust-remote-code", + "--reasoning-parser poolside_v1", + "--tool-call-parser poolside_v1", + "--tp 8", + "--host {{HOST_IP}}", + "--port {{PORT}}", + ], + }, + { + match: { hw: "b300", variant: "default", quant: "nvfp4", strategy: "low-latency", nodes: "single" }, + verified: true, + env: [], + flags: [ + "--model-path {{MODEL_NAME}}", + "--trust-remote-code", + "--reasoning-parser poolside_v1", + "--tool-call-parser poolside_v1", + "--tp 8", + "--attention-backend trtllm_mha", + "--speculative-algorithm DFLASH", + "--speculative-draft-model-path poolside/Laguna-S-2.1-DFlash-NVFP4", + "--page-size 1", + "--mem-fraction-static 0.7", + "--host {{HOST_IP}}", + "--port {{PORT}}", + ], + }, + { + match: { hw: "b300", variant: "default", quant: "int4", strategy: "high-throughput", nodes: "single" }, + verified: true, + env: [], + flags: [ + "--model-path {{MODEL_NAME}}", + "--trust-remote-code", + "--reasoning-parser poolside_v1", + "--tool-call-parser poolside_v1", + "--tp 8", + "--host {{HOST_IP}}", + "--port {{PORT}}", + ], + }, + { + match: { hw: "b300", variant: "default", quant: "int4", strategy: "low-latency", nodes: "single" }, + verified: true, + env: [], + flags: [ + "--model-path {{MODEL_NAME}}", + "--trust-remote-code", + "--reasoning-parser poolside_v1", + "--tool-call-parser poolside_v1", + "--tp 8", + "--attention-backend trtllm_mha", + "--speculative-algorithm DFLASH", + "--speculative-draft-model-path poolside/Laguna-S-2.1-DFlash-INT4", + "--page-size 1", + "--mem-fraction-static 0.7", + "--host {{HOST_IP}}", + "--port {{PORT}}", + ], + }, + + // ══════════════ GB300 (4-GPU single node, tp 4) ══════════════ + + { + match: { hw: "gb300", variant: "default", quant: "bf16", strategy: "high-throughput", nodes: "single" }, + verified: true, + env: [], + flags: [ + "--model-path {{MODEL_NAME}}", + "--trust-remote-code", + "--reasoning-parser poolside_v1", + "--tool-call-parser poolside_v1", + "--tp 4", + "--host {{HOST_IP}}", + "--port {{PORT}}", + ], + }, + { + match: { hw: "gb300", variant: "default", quant: "bf16", strategy: "low-latency", nodes: "single" }, + verified: true, + env: [], + flags: [ + "--model-path {{MODEL_NAME}}", + "--trust-remote-code", + "--reasoning-parser poolside_v1", + "--tool-call-parser poolside_v1", + "--tp 4", + "--attention-backend trtllm_mha", + "--speculative-algorithm DFLASH", + "--speculative-draft-model-path poolside/Laguna-S-2.1-DFlash", + "--page-size 1", + "--mem-fraction-static 0.7", + "--host {{HOST_IP}}", + "--port {{PORT}}", + ], + }, + { + match: { hw: "gb300", variant: "default", quant: "fp8", strategy: "high-throughput", nodes: "single" }, + verified: true, + env: ["SGLANG_SHARED_EXPERT_TP1=1"], + flags: [ + "--model-path {{MODEL_NAME}}", + "--trust-remote-code", + "--reasoning-parser poolside_v1", + "--tool-call-parser poolside_v1", + "--tp 4", + "--host {{HOST_IP}}", + "--port {{PORT}}", + ], + }, + { + match: { hw: "gb300", variant: "default", quant: "fp8", strategy: "low-latency", nodes: "single" }, + verified: true, + env: ["SGLANG_SHARED_EXPERT_TP1=1"], + flags: [ + "--model-path {{MODEL_NAME}}", + "--trust-remote-code", + "--reasoning-parser poolside_v1", + "--tool-call-parser poolside_v1", + "--tp 4", + "--attention-backend trtllm_mha", + "--speculative-algorithm DFLASH", + "--speculative-draft-model-path poolside/Laguna-S-2.1-DFlash-FP8", + "--page-size 1", + "--mem-fraction-static 0.7", + "--host {{HOST_IP}}", + "--port {{PORT}}", + ], + }, + { + match: { hw: "gb300", variant: "default", quant: "nvfp4", strategy: "high-throughput", nodes: "single" }, + verified: true, + env: [], + flags: [ + "--model-path {{MODEL_NAME}}", + "--trust-remote-code", + "--reasoning-parser poolside_v1", + "--tool-call-parser poolside_v1", + "--tp 4", + "--host {{HOST_IP}}", + "--port {{PORT}}", + ], + }, + { + match: { hw: "gb300", variant: "default", quant: "nvfp4", strategy: "low-latency", nodes: "single" }, + verified: true, + env: [], + flags: [ + "--model-path {{MODEL_NAME}}", + "--trust-remote-code", + "--reasoning-parser poolside_v1", + "--tool-call-parser poolside_v1", + "--tp 4", + "--attention-backend trtllm_mha", + "--speculative-algorithm DFLASH", + "--speculative-draft-model-path poolside/Laguna-S-2.1-DFlash-NVFP4", + "--page-size 1", + "--mem-fraction-static 0.7", + "--host {{HOST_IP}}", + "--port {{PORT}}", + ], + }, + { + match: { hw: "gb300", variant: "default", quant: "int4", strategy: "high-throughput", nodes: "single" }, + verified: true, + env: [], + flags: [ + "--model-path {{MODEL_NAME}}", + "--trust-remote-code", + "--reasoning-parser poolside_v1", + "--tool-call-parser poolside_v1", + "--tp 4", + "--host {{HOST_IP}}", + "--port {{PORT}}", + ], + }, + { + match: { hw: "gb300", variant: "default", quant: "int4", strategy: "low-latency", nodes: "single" }, + verified: true, + env: [], + flags: [ + "--model-path {{MODEL_NAME}}", + "--trust-remote-code", + "--reasoning-parser poolside_v1", + "--tool-call-parser poolside_v1", + "--tp 4", + "--attention-backend trtllm_mha", + "--speculative-algorithm DFLASH", + "--speculative-draft-model-path poolside/Laguna-S-2.1-DFlash-INT4", + "--page-size 1", + "--mem-fraction-static 0.7", + "--host {{HOST_IP}}", + "--port {{PORT}}", + ], + }, + ], +};