localairecipes launch configs for local inference

Leaderboard

Refine

Source:

Model
Quant
Storage requirement
Feature tags

NVIDIA CMP 170HX 64GB ×6 (Ampere)

RecipeAA intelC1 decodeC4 decodeC8 decodePrefillTTFTEvidenceVotes
DeepSeek-V4.1-Flash SM80 DeepInfer kernel port 6x CMP 170HX DSpark K5 #139160.8 tok/s170.5 tok/s—4506.8 tok/s10700 msreported0

NVIDIA RTX 3090 24GB ×2 (Ampere)

RecipeAA intelC1 decodeC4 decodeC8 decodePrefillTTFTEvidenceVotes
DeepSeek-V4.1-Flash-1017 dsx 2.6-bit GPTQ 2x RTX 3090 (Engram) #14526 tok/s——170 tok/s—reported0
GLM-5.3-Flash EXL3 3.05bpw on 2x RTX 3090 — vLLM + ExLlamaV3 CPU expert offload, 256K (DominikBucko)4250.4 tok/s——1155 tok/s235260 msreported0
Qwen3.8-Flash-Next W4A16-FP8PLE on 2x RTX 3090 — TP2, expert-offload, MTP3 (DominikBucko)40109.8 tok/s——3029 tok/s90800 msreported0
Qwen3.8-27B AWQ-INT4 + DSpark on RTX 3090s — SGLang TP1/TP2/TP434136.8 tok/s103.8 tok/s60.5 tok/s1615.8 tok/s98880 msreported0

NVIDIA RTX 3090 24GB ×1 (Ampere)

RecipeAA intelC1 decodeC4 decodeC8 decodePrefillTTFTEvidenceVotes
GLM-5.3-Flash EXL3 3.05bpw on 1x RTX 3090 — RAM/NVMe expert tiering (near-exact) #14228.2 tok/s——1015 tok/s—reported0
Qwen3.8-Flash-Next W4A16-FP8PLE on 1x RTX 3090 — CPU cold experts, MTP3, 128K (DominikBucko)4050.3 tok/s——2141 tok/s62200 msreported0
Qwen3.8-27B AWQ-INT4 + DSpark on RTX 3090s — SGLang TP1/TP2/TP434136.8 tok/s103.8 tok/s60.5 tok/s1615.8 tok/s98880 msreported0

AMD Radeon AI Pro R9700 32GB ×1

RecipeAA intelC1 decodeC4 decodeC8 decodePrefillTTFTEvidenceVotes
Qwen3.8-27B MXFP4 DFlash2 1x Radeon AI PRO R9700 WSL2 radiance #134145 tok/s——2762 tok/s12580 msreported0

Machine-readable: /api/v1/leaderboard.json (these boards) · /api/v1/scores.json (per-recipe medians).