{ "recipe": "R4_ds4class", "payload_bytes": 12718527024, "payload_gib": 11.84505133330822, "recipe_table": { "experts_gate": [ "q", 2, 128 ], "experts_up": [ "q", 2, 128 ], "experts_down": [ "q", 2, 128 ], "experts_gate_mtp": [ "q", 2, 128 ], "experts_up_mtp": [ "q", 2, 128 ], "experts_down_mtp": [ "q", 2, 128 ], "shared_gate": [ "q", 8, 64 ], "shared_up": [ "q", 8, 64 ], "shared_down": [ "q", 8, 64 ], "shared_gate_mtp": [ "q", 8, 64 ], "shared_up_mtp": [ "q", 8, 64 ], "shared_down_mtp": [ "q", 8, 64 ], "dense_ffn": [ "q", 8, 64 ], "router": [ "f32" ], "dsa_attn": [ "q", 8, 64 ], "dsa_norm": [ "f32" ], "indexer": [ "bf16" ], "kv_b_q8": [ "q", 8, 64 ], "kda_proj": [ "bf16" ], "kda_conv": [ "f32" ], "kda_aux": [ "f32" ], "mhc": [ "bf16" ], "norm": [ "f32" ], "norm_final": [ "f32" ], "embed": [ "q", 8, 64 ], "lm_head": [ "q", 8, 64 ], "vision": [ "bf16" ], "mtp_glue": [ "bf16" ] }, "source": "zai-org/GLM-5.3-Flash (official FP8, block e4m3 128x128)" }