{"model": "transformer inference roofline / serving model (educational)", "inputs": {"peak_tflops": 990.0, "hbm_bw_gbs": 3350.0, "hbm_cap_gb": 80.0, "params_b": 70.0, "hidden_dim": 8192.0, "layers": 80, "n_heads": 64, "n_kv_heads": 8, "seq_len": 4096.0, "gen_tokens": 256.0, "batch": 64, "weight_bytes": 2.0, "kv_bytes": 2.0}, "outputs": {"decode_tokens_s": 949.1, "per_user_tokens_s": 14.8, "per_token_latency_ms": 67.433, "decode_bottleneck": "memory-bound", "ttft_ms": 39914.05, "prefill_tokens_s": 6567.7, "prefill_bottleneck": "compute-bound", "e2e_latency_ms": 57176.8, "arithmetic_intensity": 42.71, "ridge_point": 295.52, "compute_utilization_pct": 14.5, "bandwidth_utilization_pct": 100.0, "weights_gb": 140.0, "kv_cache_gb": 85.9, "total_memory_gb": 225.9, "fits_in_hbm": false, "efficient_batch": 128}, "profile": {"batch_curve": [[1, 23.7, "memory"], [2, 46.96, "memory"], [3, 69.78, "memory"], [4, 92.18, "memory"], [5, 114.17, "memory"], [6, 135.76, "memory"], [7, 156.97, "memory"], [8, 177.79, "memory"], [9, 198.25, "memory"], [10, 218.35, "memory"], [11, 238.1, "memory"], [12, 257.52, "memory"], [13, 276.6, "memory"], [14, 295.36, "memory"], [15, 313.8, "memory"], [16, 331.94, "memory"], [17, 349.78, "memory"], [18, 367.33, "memory"], [19, 384.59, "memory"], [20, 401.57, "memory"], [21, 418.29, "memory"], [22, 434.74, "memory"], [23, 450.93, "memory"], [24, 466.87, "memory"], [25, 482.56, "memory"], [26, 498.01, "memory"], [27, 513.22, "memory"], [28, 528.21, "memory"], [29, 542.97, "memory"], [30, 557.51, "memory"], [31, 571.84, "memory"], [32, 585.95, "memory"], [33, 599.86, "memory"], [34, 613.57, "memory"], [35, 627.09, "memory"], [36, 640.4, "memory"], [37, 653.54, "memory"], [38, 666.48, "memory"], [39, 679.25, "memory"], [40, 691.84, "memory"], [41, 704.25, "memory"], [42, 716.5, "memory"], [43, 728.58, "memory"], [44, 740.5, "memory"], [45, 752.25, "memory"], [46, 763.85, "memory"], [47, 775.3, "memory"], [48, 786.6, "memory"], [49, 797.75, "memory"], [50, 808.75, "memory"], [51, 819.62, "memory"], [52, 830.34, "memory"], [53, 840.93, "memory"], [54, 851.38, "memory"], [55, 861.71, "memory"], [56, 871.9, "memory"], [57, 881.97, "memory"], [58, 891.91, "memory"], [59, 901.74, "memory"], [60, 911.44, "memory"], [61, 921.02, "memory"], [62, 930.49, "memory"], [63, 939.85, "memory"], [64, 949.1, "memory"], [65, 958.23, "memory"], [66, 967.26, "memory"], [67, 976.18, "memory"], [68, 985.0, "memory"], [69, 993.72, "memory"], [70, 1002.34, "memory"], [71, 1010.86, "memory"], [72, 1019.28, "memory"], [73, 1027.61, "memory"], [74, 1035.85, "memory"], [75, 1043.99, "memory"], [76, 1052.04, "memory"], [77, 1060.01, "memory"], [78, 1067.88, "memory"], [79, 1075.67, "memory"], [80, 1083.38, "memory"], [81, 1091.0, "memory"], [82, 1098.54, "memory"], [83, 1106.0, "memory"], [84, 1113.38, "memory"], [85, 1120.69, "memory"], [86, 1127.91, "memory"], [87, 1135.07, "memory"], [88, 1142.14, "memory"], [89, 1149.14, "memory"], [90, 1156.08, "memory"], [91, 1162.94, "memory"], [92, 1169.73, "memory"], [93, 1176.45, "memory"], [94, 1183.1, "memory"], [95, 1189.69, "memory"], [96, 1196.21, "memory"], [97, 1202.67, "memory"], [98, 1209.06, "memory"], [99, 1215.39, "memory"], [100, 1221.66, "memory"], [101, 1227.86, "memory"], [102, 1234.01, "memory"], [103, 1240.1, "memory"], [104, 1246.13, "memory"], [105, 1252.1, "memory"], [106, 1258.01, "memory"], [107, 1263.87, "memory"], [108, 1269.67, "memory"], [109, 1275.42, "memory"], [110, 1281.12, "memory"], [111, 1286.76, "memory"], [112, 1292.35, "memory"], [113, 1297.89, "memory"], [114, 1303.38, "memory"], [115, 1308.81, "memory"], [116, 1314.2, "memory"], [117, 1319.54, "memory"], [118, 1324.83, "memory"], [119, 1330.08, "memory"], [120, 1335.28, "memory"], [121, 1340.43, "memory"], [122, 1345.53, "memory"], [123, 1350.59, "memory"], [124, 1355.61, "memory"], [125, 1360.58, "memory"], [126, 1365.51, "memory"], [127, 1370.4, "memory"], [128, 1375.25, "memory"]], "selected_batch": 64, "knee_batch": 128}, "steps": 128, "node": "l4", "compute_ms": 0.4}