Recipes
Newest
6most recently addedAgents-A1 35B-A3B — FP8
InternScience's agentic-RL Agents-A1 (35B MoE, ~3B active) in the vendor's FP8 on one DGX Spark, serving the full 262,144-token context at 38.8 tok/s single-stream.
Laguna-M.1 NVFP4 (poolside)
poolside's agentic-coding Laguna-M.1 (226B MoE, ~22B active) in NVFP4 on two DGX Sparks (TP2) — 12.5 tok/s single-stream at 196,608-token context.
AREX-Base 122B-A10B — Q4_K_M
BAAI's AREX-Base deep-research agent (122B MoE, ~10B active) in Q4_K_M on one DGX Spark — 23.98 tok/s single-stream, at a measured 250,223-token context.
LFM2.5 8B-A1B (Liquid AI) — BF16
Liquid AI's LFM2.5 8B-A1B (8.4B MoE, ~1B active) unquantized in BF16 on one DGX Spark, serving the full 128,000-token context at 57.5 tok/s single-stream.
Qwen AgentWorld 35B-A3B — NVFP4
Qwen's agentic AgentWorld 35B-A3B (35B MoE, ~3B active) in NVFP4 on one DGX Spark, serving the full 262,144-token context at 83.1 tok/s single-stream with Qwen3.6's MTP head grafted in.
Ornith 1.0 9B — BF16
Deep Reinforce's agentic-coding Ornith 1.0 9B (9B dense) unquantized in BF16 on one DGX Spark, serving the full 262,144-token context at 12.6 tok/s single-stream.
1 Spark
23Agents-A1 35B-A3B — FP8
InternScience's agentic-RL Agents-A1 (35B MoE, ~3B active) in the vendor's FP8 on one DGX Spark, serving the full 262,144-token context at 38.8 tok/s single-stream.
KAT-Coder V2.5-Dev NVFP4
Kwaipilot's agentic-coding KAT-Coder V2.5-Dev (35B MoE, ~3B active) in full NVFP4 on one DGX Spark, serving the full 262,144-token context at 82 tok/s single-stream decode — a 27% gain over the same config with no draft, from grafting Qwen3.6-35B-A3B's multi-token-prediction head into a checkpoint that shipped without one. Speculative decoding is verified, so the draft costs nothing in output quality, and the whole working set is 38 of 114 usable GiB.
Laguna-S 2.1 NVFP4
poolside's agentic-coding Laguna-S 2.1 (118B MoE, ~8B active) in NVFP4 on one DGX Spark, with our W4A16 Sparkulator quant of the matching NVFP4 DFlash speculator — 45.9 tok/s single-stream at the full 262K context, +7% over poolside's BF16 draft at identical acceptance.
Nemotron 3 Nano 30B-A3B — NVFP4
NVIDIA's Nemotron 3 Nano 30B-A3B (31.6B hybrid Mamba-MoE, ~3B active) in NVFP4 on one DGX Spark, serving the full 262,144-token context — 61.2 tok/s single-stream, and still 56.0 tok/s on a 131K-token prompt.
Laguna-XS 2.1 NVFP4
Serve poolside's agentic-coding Laguna-XS 2.1 (33B MoE, ~3B active) on one DGX Spark with the NVFP4 experts and a W4A16 Sparkulator quant of the checkpoint's own DFlash speculator: 133.6 tok/s single-stream on code — 3.2x the no-draft baseline — with the full 262K context still fitting 1.5x over. Needs vLLM 0.25.1 (the release where the Laguna DFlash draft landed) and two non-obvious tuning calls: k=6, not k=1, and --max-num-seqs 2, not 4.
Qwen3.6 35B-A3B NVFP4 (Unsloth Fast)
Serve the full 262,144-token context of Qwen3.6 35B-A3B on ONE DGX Spark at 106 tok/s single-stream, using the checkpoint's own MTP head for speculative decode at k=3 — a 57% gain over the same config with no draft. Nothing needs patching and nothing is tight: the whole working set is 36 of 114 usable GiB.
Laguna-XS 2.1 Q4_K_M (llama.cpp)
Serve poolside's agentic-coding Laguna-XS 2.1 (33B MoE, ~3B active) on one DGX Spark as a 20.3 GB Q4_K_M GGUF: 90.4 tok/s single-stream with no draft, 122.7 with the DFlash speculator, and the model's full native 262,144-token window open the whole time — which, measured, costs about 1 GiB and zero decode speed. Needs a llama.cpp build carrying the laguna architecture port; stock llama.cpp cannot load this model at all.
Qwen3.6 35B-A3B — FP8
Qwen's own FP8 build of Qwen3.6 35B-A3B serves the full 262,144-token context on one DGX Spark in about 41 GiB and, unlike the NVFP4A16 build, it gets a real FP8 kernel path — vLLM picks the TRITON FP8 MoE backend, not the MARLIN fallback. It still decodes slower: 38.3 tok/s against 42.6 for NVFP4A16 and 106.5 for Unsloth's NVFP4-Fast. The reason is bytes, not kernels — the FP8 export carries 30.1 GiB of expert planes against roughly 17.5 — and like every official Qwen3.6 export it ships no MTP head.
1Kimi Linear 48B-A3B — BF16
Moonshot's Kimi Linear 48B-A3B (49B MoE, ~2.7B active) in BF16 on one DGX Spark, at a 98,304-token context — 29.4 tok/s single-stream decode, with the KV cache costing only 8.2 KB per token.
Nemotron 3 Nano 30B-A3B — BF16
The unquantized BF16 build of NVIDIA's Nemotron 3 Nano 30B-A3B on one DGX Spark. It serves the same full 262,144-token context as the NVFP4 recipe — but it costs 68 GiB instead of 28.3, and decodes at 28.9 tok/s instead of 61.2. This is the page that tells you what NVFP4 is actually buying.
Qwen3.6 27B NVFP4 (NVIDIA ModelOpt)
Qwen3.6 27B — a dense 27B with hybrid GDN/full attention and a vision tower — in mixed NVFP4 on one DGX Spark, with the checkpoint's own MTP head speculating 5 tokens ahead: 34.5 tok/s single-stream, 2.8× the no-draft baseline measured in the same session, at the full 262,144-token context.
Qwen3.6 35B-A3B — NVIDIA NVFP4
NVIDIA's official NVFP4 build of Qwen3.6 35B-A3B serves the full 262,144-token context on one DGX Spark — but it decodes at 42.6 tok/s against 106.5 for Unsloth's NVFP4-Fast build of the same base model. Two structural reasons, both measured here: it is weight-only NVFP4, so the FP4 tensor-core MoE kernels refuse it and vLLM falls back to MARLIN; and the export ships without the MTP head, so it cannot speculative-decode at all.
Qwen AgentWorld 35B-A3B — BF16
Qwen's agentic AgentWorld 35B-A3B (35B MoE, ~3B active) unquantized in BF16 on one DGX Spark, serving the full 262,144-token context at 30.4 tok/s single-stream.
Ornith 1.0 35B — BF16
Deep Reinforce's agentic-coding Ornith 1.0 35B (35B MoE, ~3B active) unquantized in BF16 on one DGX Spark, serving the full 262,144-token context at 30.3 tok/s single-stream.
Qwen3-Coder-Next NVFP4
Qwen's agentic-coding Qwen3-Coder-Next (80B MoE, ~3B active) in NVFP4 on one DGX Spark, with a 1-layer EAGLE3 speculator — 72.0 tok/s single-stream, 1.18× the no-draft baseline, at the full 262,144-token context.
T-Search 35B-A3B NVFP4
T-Bank's agentic-search T-Search (36B MoE, ~3B active) in NVFP4 on one DGX Spark, with the checkpoint's own MTP head speculating 3 tokens ahead — 62.0 tok/s single-stream, 1.68x the no-draft baseline, at the full 262,144-token context. Re-verified unchanged on vLLM 0.26.0.
BTL-3 (Qwen3.6-27B) — BF16
badtheorylabs' BTL-3 agentic coder (a LoRA on Qwen3.6-27B, ~27B dense) in BF16 on one DGX Spark, with the base model's native MTP head driving speculative decode — 12.8 tok/s single-stream, 2.9x the no-draft baseline, at the full 262K context.
Ornith 1.0 35B — FP8
Deep Reinforce's agentic-coding Ornith 1.0 35B (35B MoE, ~3B active) in vendor FP8 on one DGX Spark, serving the full 262,144-token context at 37.7 tok/s single-stream.
Ornith 1.0 9B — BF16
Deep Reinforce's agentic-coding Ornith 1.0 9B (9B dense) unquantized in BF16 on one DGX Spark, serving the full 262,144-token context at 12.6 tok/s single-stream.
Gemma 4 26B-A4B — FP8-Dynamic
Gemma 4 26B-A4B at FP8 serves its full 262,144-token context on a single DGX Spark in about 34 GiB — a quarter of the box — because 25 of its 30 layers are sliding-window attention and only 5 keep a global KV cache. Pair it with Google's 0.42B assistant draft and decode goes from 38.6 to 66.8 tok/s at a 2K prompt; quantize that draft to W4A16 (Sparkulator-Gemma-4-26B-A4B) and it goes to 73.9, for free — speculative decoding is verified, so a lossier draft cannot change a single output token. The draft must be run on vLLM 0.24.0: 0.26.0 crashes on it.
Qwen AgentWorld 35B-A3B — NVFP4
Qwen's agentic AgentWorld 35B-A3B (35B MoE, ~3B active) in NVFP4 on one DGX Spark, serving the full 262,144-token context at 83.1 tok/s single-stream with Qwen3.6's MTP head grafted in.
LFM2.5 8B-A1B (Liquid AI) — BF16
Liquid AI's LFM2.5 8B-A1B (8.4B MoE, ~1B active) unquantized in BF16 on one DGX Spark, serving the full 128,000-token context at 57.5 tok/s single-stream.
AREX-Base 122B-A10B — Q4_K_M
BAAI's AREX-Base deep-research agent (122B MoE, ~10B active) in Q4_K_M on one DGX Spark — 23.98 tok/s single-stream, at a measured 250,223-token context.
2 Sparks
6DeepSeek V4 Flash DSpark
Serve the 284B/13B DeepSeek V4 Flash DSpark checkpoint across two DGX Sparks at its full 1M-token context, with FP4 experts, an NVFP4 MLA KV cache, and DSpark speculative decoding at k=5 — 42.0 tok/s single-stream on prose and 76.0 on code.
2GLM-5.2 — 2-bit MoE
GLM-5.2 (753B MoE) in 2-bit experts + NVFP4 attention on two DGX Sparks (TP2), with a W4A16 dspark speculator at K=3 — 25.8 tok/s single-stream at 96K context.
Laguna-M.1 NVFP4 (poolside)
poolside's agentic-coding Laguna-M.1 (226B MoE, ~22B active) in NVFP4 on two DGX Sparks (TP2) — 12.5 tok/s single-stream at 196,608-token context.
2Hy3 NVFP4
Serve the full Hunyuan Hy3 (295B MoE, 21B active) across two DGX Sparks with NVFP4 expert planes, BF16 attention, and the checkpoint's native MTP head driving speculative decode — 19.5 tok/s single-stream at 160K context. Needs the marlin MoE backend on sm_121 and a one-line vLLM patch for the MTP head; both are covered below.
Laguna-S 2.1 FP8
poolside's agentic-coding Laguna-S 2.1 (118B MoE, ~8B active) in FP8 on two DGX Sparks (TP2), with a self-quantized W4A16 DFlash speculator — 60.9 tok/s single-stream (2.3x the no-draft baseline) at the full 262K context.
2Solar-Open2 250B NVFP4
Serve Upstage's Solar-Open2 (250B-class hybrid linear-attention MoE, 320 experts / top-8, ~8B active) in Nota's NVFP4 build across two DGX Sparks (TP2) — 10.2 tok/s single-stream decode, serving the full 1M native context (NoPE, no rope scaling), decode-bound by the BF16 attention the quant leaves untouched. Runs on a hand-written out-of-tree vLLM model plugin (no released engine implements the architecture).