Skip to content

EN

Local AI · Creator

r0b0tlab

42 published recipes.

Recipes

1× DGX Spark

MiniCPM-V-4.7-35B-A3B EXL3 6.08 bpw with ExLlamaV3 on 1× DGX Spark (text, 262k context)

59.4tok/sEveryday

1 request, chat prompt (20 prompts, thinking off), no drafting (default profile) Source

Intelligenceno independent value

Custom kernel required

EXL3 6.08 bpw (28,3 GB), FP16 KV; nur Textpfad (Vision-Tower wird von ExLlamaV3 nicht ausgeführt)ExLlamaV3 1.5.0 (Fork r0b0tlab/exllamav3-r0b0t, Branch minicpm-v4.7-dspark @ 5ddae7d12448), torch 2.14.0+cu130Weights Repo Updated Oct 14, 2026

1× RTX 3090

Qwen3.8 27B EXL3 4.00 bpw with ExLlamaV3 on 1× RTX 3090

42.8tok/sEveryday

1 request, prose prompt, no speculative decoding Source

Intelligence34with thinking no thinking 20

EXL3 4.00 bpw (15.4 GiB)ExLlamaV3Weights Repo Updated Sep 20, 2026

2× DGX Spark

DeepSeek-V4-Flash-Vision-Exp FP8 with vLLM on 2× DGX Spark

32.5tok/sEveryday

1 request, prose prompt, with DSpark (adaptive verification) (per recipe) Source

Intelligenceno independent value

Experimental
Custom kernel required

offizielle Gewichte + FP8 KVvLLMWeights Repo Updated Sep 3, 2026

1× DGX Spark

Qwen3.8-Flash-Next 180B NVFP4 with SGLang on 1× DGX Spark

26.4tok/sEveryday

1 request, prose prompt, with MTP (NEXTN) (per recipe) Source

Intelligence40with thinking

Custom kernel required
License unclear

NVFP4 W4A4 Experts + FP8 MTPSGLang 84cf99860e30 (gepatcht)Weights Repo Updated Sep 13, 2026

1× DGX Spark

GLM-5.3-Flash EXL3 2.25 bpw with ExLlamaV3 on 1× DGX Spark (256k context)

20.6tok/sEveryday

1 request, prose prompt, with DFlash2 K=5 (per recipe) Source

Intelligence42with thinkingOriginal model's value. Heavily compressed (~2.3 bit), intelligence is likely much lower.

≤ 3 bit: quantization may cost quality
Custom kernel required
Non-commercial

EXL3 2.25 bpw (Drafter EXL3 3.00 bpw), FP16 KVExLlamaV3 v1.5.2 + TabbyAPI 816c321 (ghcr.io/r0b0tlab/glm53-flash-exl3-exllamav3-gb10:b25c467)Weights Repo Updated Sep 28, 2026

1× DGX Spark

GLM-5.3-Flash EXL3 2.25 bpw with vLLM on 1× DGX Spark

19.5tok/sEveryday

1 request, prose prompt, with DFlash2 (per recipe) Source

Intelligence42with thinkingOriginal model's value. Heavily compressed (~2.3 bit), intelligence is likely much lower.

≤ 3 bit: quantization may cost quality
Experimental
Custom kernel required
Non-commercial

EXL3 2.25 bpw (Drafter EXL3 3.00 bpw)vLLM v0.30.0rc1 @ a00a3544b93eWeights Repo Updated Sep 27, 2026

2× DGX Spark

GLM-5.3-Flash NVFP4 with vLLM on 2× DGX Spark (127k context, with DFlash2)

17.9tok/sEveryday

1 request, prose prompt, with DFlash2 Source

Intelligence42with thinking

Custom kernel required
Non-commercial

NVFP4 (ModelOpt, offiziell) + BF16 KVvLLM 0.28.1rc1.dev580Weights Repo Updated Sep 11, 2026

1× DGX Spark

Nemotron-3.5-Lightning 30B NVFP4 with vLLM on 1× DGX Spark

99.5tok/sMixed

1 request, mixed prompt set, with MTP (per recipe) Source

Intelligence13with thinking

License unclear

NVFP4 + FP8 KVvLLM 0.23.1rc1.dev1327+gf25953cc5Weights Repo Updated Aug 11, 2026

1× DGX Spark

Qwen3.6-35B-A3B NVFP4 with vLLM on 1× DGX Spark (256k context)

93.1tok/sMixed

1 request, mixed prompt set, with MTP (per recipe) Source

Intelligence18with thinking no thinking 15

NVFP4 + FP8 KVvLLM 0.25.0 (702f4814)Weights Repo Updated Jul 15, 2026

2× DGX Spark

DeepSeek-V4-Flash with vLLM on 2× DGX Spark (1024k context)

86.9tok/sMixed

1 request, mixed prompt set, with DSpark (per recipe) Source

Intelligence34with thinking

Custom kernel required

deepseek_v4_fp8 Gewichte + nvfp4_ds_mla KVvLLM 0.26.0+dspark.sm121.3Weights Repo Updated Aug 3, 2026

1× DGX Spark

Ornith-1.5-35B-A3B NVFP4 with SGLang on 1× DGX Spark

82.2tok/sMixed

1 request, mixed prompt set, with MTP (EAGLE/nextn) (per recipe) Source

Intelligenceno independent value

NVFP4 W4A16 (weight-only, ModelOpt w4_nvfp4), MTP-Kopf BF16SGLang 0.5.6.post3.dev9218+g5a7b26c63Weights Repo Updated Aug 24, 2026

1× DGX Spark

Qwen3.6-35B-A3B NVFP4 with vLLM on 1× DGX Spark (64k context)

79.7tok/sMixed

1 request, mixed prompt set, with MTP (per recipe) Source

Intelligence18with thinking no thinking 15

NVFP4 experts + FP8 dense (compressed-tensors) + FP8 KVvLLM 0.24.1-devWeights Repo Updated Jul 10, 2026

1× DGX Spark

Nemotron-3.5-Lightning 30B NVFP4 with SGLang on 1× DGX Spark

66.7tok/sMixed

1 request, mixed prompt set, with DFlash Source

Intelligence13with thinking

License unclear

NVFP4 (W4A16 Marlin MoE) + FP8 KVSGLangWeights Repo Updated Aug 21, 2026

1× DGX Spark

XYZ-Aquila-mini NVFP4 with vLLM on 1× DGX Spark

64.6tok/sMixed

1 request, mixed prompt set, no speculative decoding Source

Intelligenceno independent value

NVFP4 (modelopt_mixed) + FP8 KVvLLM 0.25.0 (702f4814)Weights Repo Updated Jul 29, 2026

2× DGX Spark

Qwen3.8-Flash-Next 180B NVFP4 with SGLang on 2× DGX Spark

62.1tok/sMixed

1 request, mixed prompt set, with MTP (NEXTN) (per recipe) Source

Intelligence40with thinking

License unclear

NVFP4 W4A16 (ModelOpt) + FP8 E4M3 KV, MTP BF16SGLang 7d2e8fcb9a76Weights Repo Updated Sep 1, 2026

1× DGX Spark

Qwen3.6-35B-A3B NVFP4 with SGLang on 1× DGX Spark

57.5tok/sMixed

1 request, mixed prompt set, with MTP Source

Intelligence18with thinking no thinking 15

NVFP4 (compressed-tensors)SGLangWeights Repo Updated May 19, 2026

1× DGX Spark

Qwen3.6-35B-A3B NVFP4 with vLLM on 1× DGX Spark (4k context)

45.4tok/sMixed

1 request, mixed prompt set, no speculative decoding Source

Intelligence18with thinking no thinking 15

NVFP4 (Marlin backend)vLLM 0.20.1Weights Repo Updated May 18, 2026

1× DGX Spark

Laguna-XS-2.1 33B NVFP4 with vLLM on 1× DGX Spark

43.8tok/sMixed

1 request, mixed prompt set, no speculative decoding Source

Intelligenceno independent value

NVFP4 (compressed-tensors) + FP8 KVvLLM v0.24.0Weights Repo Updated Jul 4, 2026

1× DGX Spark

Ling-3.0-flash-VL 124B NVFP4 with vLLM on 1× DGX Spark

43.5tok/sMixed

1 request, mixed prompt set, no speculative decoding Source

Intelligence25with thinking

Custom kernel required

NVFP4 + FP8 mixed (ModelOpt 0.48) + FP8 KVvLLM vllm-ling-v3 @ e2e5751Weights Repo Updated Sep 9, 2026

1× RTX 3090

Qwen3.8-Flash-Next 125B-A6B EXL3 2.50 bpw with ExLlamaV3 on 1× RTX 3090

38.6tok/sMixed

1 request, mixed prompt set, with MTP Source

Intelligence40with thinking

≤ 3 bit: quantization may cost quality
Partly in system RAM
License unclear

EXL3 2.50 bpw (41.5 GiB) + 3-bit n-gram TabelleExLlamaV3Weights Repo Updated Sep 17, 2026

2× DGX Spark

DeepSeek-V4-Flash MXFP4 with vLLM on 2× DGX Spark

36.1tok/sMixed

1 request, mixed prompt set, with MTP (per recipe) Source

Intelligence24with thinking no thinking 19

Experimental
Custom kernel required

FP8 (offiziell, block-scaled) + MXFP4 MoEvLLMWeights Repo Updated Jun 16, 2026

1× DGX SparkNex-N2-mini (Qwen3.5-MoE)

Nex-N2-mini 35B NVFP4 with vLLM on 1× DGX Spark

33.4tok/sMixed

1 request, mixed prompt set, no speculative decoding Source

Intelligenceno independent value

NVFP4 (ModelOpt 0.44, MLP-only) + FP8 KVvLLM v0.22.0Weights Repo Updated Jun 9, 2026

1× DGX Spark

Gemma-4-26B-A4B NVFP4 with vLLM on 1× DGX Spark

29.6tok/sMixed

1 request, mixed prompt set, no speculative decoding Source

Intelligence17with thinking no thinking 13

NVFP4 (ModelOpt) + FP8 KVvLLMWeights Repo Updated May 19, 2026

1× RTX 3090

Qwen3.8-27B EXL3 4.00 bpw with buun-llama-cpp on 1× RTX 3090

27.8tok/sMixed

1 request, mixed prompt set, context 150000, with DFlash2 (per recipe) Source

Intelligence34with thinking no thinking 20

Custom kernel required

EXL3 4.00 bpw (Target + DFlash2-Draft), VBR-KV turbo8/turbo3buun-llama-cppWeights Repo Updated Sep 20, 2026

1× DGX Spark

Qwen3.8-27B NVFP4 with vLLM on 1× DGX Spark (r0b0tlab)

27.8tok/sMixed

1 request, mixed prompt set, with MTP Source

Intelligence34with thinking no thinking 20

License unclear

NVFP4 (+ BF16-MTP-Kopf)vLLM v0.27.2rc0-sm121 (7f7a32c)Weights Repo Updated Aug 20, 2026

1× DGX Spark

Qwen3.6-27B NVFP4 with vLLM on 1× DGX Spark

26.6tok/sMixed

1 request, mixed prompt set, with MTP Source

Intelligence21with thinking no thinking 20

NVFP4 + FP8 KVvLLM v0.25.1Weights Repo Updated Jul 20, 2026

2× DGX Spark

MiniMax-M2.7 NVFP4 with vLLM on 2× DGX Spark

24.1tok/sMixed

1 request, mixed prompt set, no speculative decoding Source

Intelligence23with thinking

Custom kernel required
License unclear

NVFP4 (ModelOpt)vLLM v0.21.1 (source build)Weights Repo Updated May 24, 2026

1× DGX Spark

Qwen3.8-27B NVFP4 with SGLang on 1× DGX Spark (r0b0tlab)

23.5tok/sMixed

1 request, mixed prompt set, with DFlash2 (per recipe) Source

Intelligence34with thinking no thinking 20

Custom kernel required
License unclear

NVFP4 (MTP-Kopf enthalten)SGLangWeights Repo Updated Aug 19, 2026

1× DGX Spark

Ling-3.0-flash NVFP4 with vLLM on 1× DGX Spark

21.4tok/sMixed

1 request, mixed prompt set, with MTP (per recipe) Source

Intelligence25with thinking

Custom kernel required

NVFP4 (ModelOpt, nur Experten) + FP8 KVvLLM vllm-ling-v3 @ 92c104112Weights Repo Updated Aug 8, 2026

1× DGX Spark

Laguna-S-2.1 NVFP4 with vLLM on 1× DGX Spark

20.6tok/sMixed

1 request, mixed prompt set, with DFlash (per recipe) Source

Intelligenceno independent value

Custom kernel required
License unclear

NVFP4 + FP8 KVvLLM v0.25.1Weights Repo Updated Jul 23, 2026

1× DGX SparkAgents-A1 (Qwen3.5-MoE)

Agents-A1 NVFP4 with vLLM on 1× DGX Spark

17.5tok/sMixed

1 request, mixed prompt set, no speculative decoding Source

Intelligenceno independent value

Experimental

NVFP4 MLP/MoE-only (ModelOpt) + FP8 KVvLLMWeights Repo Updated Jul 1, 2026

1× DGX Spark

Nemotron-3-Super-120B-A12B NVFP4 with SGLang on 1× DGX Spark

14.8tok/sMixed

1 request, mixed prompt set, with MTP Source

Intelligence13with thinking

License unclear

NVFP4 (ModelOpt)SGLangWeights Repo Updated May 19, 2026

2× DGX Spark

Step-3.7-Flash 198B NVFP4 with vLLM on 2× DGX Spark

13.2tok/sMixed

1 request, mixed prompt set, no speculative decoding Source

Intelligence20with thinking

NVFP4 (ModelOpt 0.45.0) + FP8 KVvLLMWeights Repo Updated Jun 11, 2026

2× DGX Spark

Laguna-M.1 225B NVFP4 with vLLM on 2× DGX Spark

11.4tok/sMixed

1 request, mixed prompt set, no speculative decoding Source

Intelligenceno independent value

Custom kernel required

NVFP4 + NVFP4 KVvLLMWeights Repo Updated Jun 20, 2026

4× DGX Spark

DeepSeek-V4.1-Flash with SGLang on 4× DGX Spark

11.2tok/sMixed

1 request, mixed prompt set, with DSpark (per recipe) Source

Intelligence40with thinking no thinking 25

Custom kernel required

offizieller Checkpoint (unquantisiert laut REPRODUCIBILITY.md: „does not … quantize the checkpoint“)SGLangWeights Repo Updated Sep 16, 2026

1× DGX Spark

Muse-Glimmer-30B NVFP4 with vLLM on 1× DGX Spark

10.3tok/sMixed

1 request, mixed prompt set, no speculative decoding Source

Intelligence18with thinking

Experimental
Custom kernel required

NVFP4 (ModelOpt 0.45, eigene Quantisierung; Gewichte nicht publiziert)vLLMWeights Repo Updated Aug 15, 2026

1× DGX Spark

Gemma-4-31B NVFP4 with vLLM on 1× DGX Spark

6.88tok/sMixed

1 request, mixed prompt set, no speculative decoding Source

Intelligence19with thinking no thinking 14

License unclear

NVFP4 (ModelOpt) + FP8 KVvLLM 0.21.0Weights Repo Updated May 19, 2026

1× DGX Spark

DiffusionGemma-26B-A4B NVFP4 with vLLM on 1× DGX Spark

146tok/sPeak

1 request, mixed prompt set Source

Intelligence10with thinking

Experimental

NVFP4 (ModelOpt) + FP8 KVvLLM day-0 DiffusionGemma-WheelWeights Repo Updated Jun 10, 2026

1× DGX Spark

VibeThinker-3B NVFP4 with vLLM on 1× DGX Spark

71.3tok/sPeak

1 request, synthetic prompt, context 2048 Source

Intelligenceno independent value

NVFP4 W4A4 (ModelOpt 0.44.0), FP8 KVvLLMWeights Repo Updated Jun 17, 2026

1× DGX SparkFastContext-1.0 (Qwen3-4B)

FastContext-1.0 NVFP4 with vLLM on 1× DGX Spark

66.3tok/sPeak

1 request, mixed prompt set Source

Intelligenceno independent value

broken

NVFP4 W4A4 (ModelOpt) + FP8 KVvLLM >=0.23Weights Repo Updated Jun 18, 2026

2× DGX Spark

Inkling-Small NVFP4 with SGLang on 2× DGX Spark

16tok/sPeak

1 request, mixed prompt set Source

Intelligence28with thinking

NVFP4 (modelopt_fp4)SGLangWeights Repo Updated Aug 7, 2026

2× DGX Spark

MiMo-V2.6-Flash NVFP4 with SGLang on 2× DGX Spark

no documented speed

Intelligenceno independent value

Custom kernel required

NVFP4 weights + kalibrierter FP8 KVSGLang nightly-dev-cu13 20260922-582389ceWeights Repo Updated Sep 30, 2026

← Back to overview