COLLECTION / RECIPES

Recipes

One artifact × hardware × engine unit. Validated rows are launch-safe; measured speed sweeps attach to each recipe as evidence.

41 records

Apple M3 Ultra 512GB1 × 512 GBGLM-5.22.56bpw-mixedmlxreferenceUnknown contextcontext20.5 tok/smeasuredcandidate mlx evidence measured

Apple M3 Ultra 512GB1 × 512 GBGLM-5.23.5bpw-mixedmlxreferenceUnknown contextcontext18.5 tok/smeasuredcandidate mlx evidence measured

Apple M3 Ultra 512GB1 × 512 GBGLM-5.2mixed-3_6bitmlxreferenceUnknown contextcontext14.8 tok/smeasuredcandidate mlx evidence measured

Apple M3 Ultra 512GB1 × 512 GBGLM-5.2MLX-2.32bpw-dynamicmlxreferenceUnknown contextcontext19.8 tok/smeasuredcandidate mlx evidence measured

Apple M3 Ultra 512GB1 × 512 GBGLM-5.2MLX-2.56bpw-dynamicmlxreferenceUnknown contextcontext19.7 tok/smeasuredcandidate mlx evidence measured

GeForce RTX 30908 × 24 GBGLM-5.1IQ1_Sllama.cppreference8Kcontext40.8 tok/smeasuredcandidate llama.cpp evidence measured

GeForce RTX 309010 × 24 GBGLM-5.2Q3_K_Mllama.cppreference256Kcontext22.1 tok/smeasuredcandidate llama.cpp evidence measured

GeForce RTX 30908 × 24 GBGLM-5.3-FlashUnsloth-Dynamic-IQ4_XSllama.cppreference128Kcontext42.1 tok/smeasuredcandidate llama.cpp evidence measured

GeForce RTX 30908 × 24 GBGLM-5.3-FlashAWQ-4bitvllmreference200Kcontext75.2 tok/smeasuredcandidate vllm evidence measured

NVIDIA B2008 × 180 GBGLM-5.2NVFP4vllmdockerUnknown contextcontext—no evidencecandidate vllm docker unmeasured

NVIDIA DGX Spark GB101 × 128 GBGLM-5.2Int4-Int8Mixvllmreference195.3Kcontext38.6 tok/smeasuredcandidate vllm evidence measured

NVIDIA DGX Spark GB104 × 128 GBGLM-5.2W4A16 + Int8Mixvllmscript308.6Kcontext618 tok/smeasuredcandidate vllm no docker measured

NVIDIA DGX Spark GB103 × 128 GBGLM-5.2NVFP4 hot experts + AQLM 2-bit cold expertsvllmscript39.1Kcontext21 tok/smeasuredcandidate vllm no docker measured

NVIDIA DGX Spark GB102 × 128 GBGLM-5.3-FlashNVFP4sglangdocker128Kcontext17.5 tok/smeasuredvalidated sglang docker measured

NVIDIA DGX Spark GB104 × 128 GBGLM-5.3-FlashNVFP4sglangdocker1024Kcontext22.7 tok/smeasuredvalidated sglang docker measured

NVIDIA DGX Spark GB102 × 128 GBGLM-5.3-Flashexl3vllmdockerUnknown contextcontext—no evidencecandidate vllm docker unmeasured

Radeon AI PRO R97001 × 32 GBGLM-5.3-FlashUD-IQ1_Mllama.cppreference32Kcontext17.4 tok/smeasuredcandidate llama.cpp evidence measured

Radeon AI PRO R97001 × 32 GBGLM-5.3-FlashUD-Q2_K_XLllama.cppreference32Kcontext16.6 tok/smeasuredcandidate llama.cpp evidence measured

RTX PRO 6000 Blackwell8 × 96 GBGLM-5.1-NVFP4NVFP4sglangreferenceUnknown contextcontext74.5 tok/smeasuredcandidate sglang evidence measured

RTX PRO 6000 Blackwell4 × 96 GBGLM-5.2UD-IQ4_XSllama.cppreference64Kcontext39.8 tok/smeasuredcandidate llama.cpp evidence measured

RTX PRO 6000 Blackwell4 × 96 GBGLM-5.2UD-Q2_K_MXFP4llama.cppreference512Kcontext47.9 tok/smeasuredcandidate llama.cpp evidence measured

RTX PRO 6000 Blackwell4 × 96 GBGLM-5.2MXFP8_NVFP4_NF3vllmreference128Kcontext36.8 tok/smeasuredcandidate vllm evidence measured

RTX PRO 6000 Blackwell4 × 96 GBGLM-5.2NVFP4vllmreference468.8Kcontext79.6 tok/smeasuredcandidate vllm evidence measured

RTX PRO 6000 Blackwell4 × 96 GBGLM-5.23.0 bpwvllmdocker256Kcontext188.1 tok/smeasuredvalidated vllm docker measured