From 89e501c5a81f038ffa391fda42b53c7cd361b03f Mon Sep 17 00:00:00 2001 From: Mick Date: Sun, 17 May 2026 23:35:59 +0800 Subject: [PATCH] [diffusion] CI: tighten selected perf baselines (#25510) --- .../multimodal_gen/runtime/server_args.py | 15 +- .../server/ascend/perf_baselines_npu.json | 5 +- .../test/server/perf_baselines.json | 970 +++++++++--------- 3 files changed, 498 insertions(+), 492 deletions(-) diff --git a/python/sglang/multimodal_gen/runtime/server_args.py b/python/sglang/multimodal_gen/runtime/server_args.py index f747a0eaf..0673e7932 100644 --- a/python/sglang/multimodal_gen/runtime/server_args.py +++ b/python/sglang/multimodal_gen/runtime/server_args.py @@ -1201,6 +1201,7 @@ class ServerArgs(DisaggArgsMixin): help="The number of warmup steps to perform for each resolution.", ) + # layerwise offload parser.add_argument( "--dit-cpu-offload", action=StoreBoolean, @@ -1233,11 +1234,8 @@ class ServerArgs(DisaggArgsMixin): default=ServerArgs.dit_offload_prefetch_size, help="The size of prefetch for dit-layerwise-offload. If the value is between 0.0 and 1.0, it is treated as a ratio of the total number of layers. If the value is >= 1, it is treated as the absolute number of layers. 0.0 means prefetch 1 layer (lowest memory). Values above 0.5 might have peak memory close to no offload but worse performance.", ) - parser.add_argument( - "--use-fsdp-inference", - action=StoreBoolean, - help="Use FSDP inference to shard DiT weights across GPUs. For single-GPU memory pressure, prefer CPU or layerwise offload.", - ) + + # offload flags parser.add_argument( "--text-encoder-cpu-offload", action=StoreBoolean, @@ -1253,6 +1251,12 @@ class ServerArgs(DisaggArgsMixin): action=StoreBoolean, help="Use CPU offload for VAE. Enable if run out of memory.", ) + + parser.add_argument( + "--use-fsdp-inference", + action=StoreBoolean, + help="Use FSDP inference to shard DiT weights across GPUs. For single-GPU memory pressure, prefer CPU or layerwise offload.", + ) parser.add_argument( "--pin-cpu-memory", action=StoreBoolean, @@ -1278,6 +1282,7 @@ class ServerArgs(DisaggArgsMixin): help="Disable autocast for denoising loop and vae decoding in pipeline sampling", ) + # quantization parser.add_argument( "--quantization", type=str, diff --git a/python/sglang/multimodal_gen/test/server/ascend/perf_baselines_npu.json b/python/sglang/multimodal_gen/test/server/ascend/perf_baselines_npu.json index dbdceba10..798d4e957 100644 --- a/python/sglang/multimodal_gen/test/server/ascend/perf_baselines_npu.json +++ b/python/sglang/multimodal_gen/test/server/ascend/perf_baselines_npu.json @@ -147,7 +147,7 @@ "per_frame_generation": null }, "denoise_step_ms": { - "0": 101.91, + "0": 330.0, "1": 303.37, "2": 548.94, "3": 542.02, @@ -200,7 +200,8 @@ }, "expected_e2e_ms": 29360.56, "expected_avg_denoise_ms": 524.67, - "expected_median_denoise_ms": 537.54 + "expected_median_denoise_ms": 537.54, + "estimated_full_test_time_s": 157.8 }, "wan2_2_t2v_14b_w8a8_8npu": { "stages_ms": { diff --git a/python/sglang/multimodal_gen/test/server/perf_baselines.json b/python/sglang/multimodal_gen/test/server/perf_baselines.json index cf224c3a8..73d3b7b2b 100644 --- a/python/sglang/multimodal_gen/test/server/perf_baselines.json +++ b/python/sglang/multimodal_gen/test/server/perf_baselines.json @@ -2,8 +2,8 @@ "metadata": { "model": "Diffusion Server", "hardware": "CI H100 80GB pool", - "description": "Reference numbers captured from CI H100 runs (2026-04-15).", - "last_updated": "2026-04-15" + "description": "Reference numbers captured from CI H100 runs (2026-05-17).", + "last_updated": "2026-05-17" }, "tolerances": { "long_term": { @@ -523,68 +523,68 @@ }, "flux_image_t2i_2_gpus": { "stages_ms": { - "InputValidationStage": 0.04, - "TextEncodingStage": 61.66, - "TimestepPreparationStage": 26.36, - "LatentPreparationStage": 0.26, - "DenoisingStage": 5299.62, - "DecodingStage": 17.87 + "InputValidationStage": 0.08, + "TextEncodingStage": 38.63, + "TimestepPreparationStage": 38.66, + "LatentPreparationStage": 0.38, + "DenoisingStage": 5000.66, + "DecodingStage": 11.98 }, "denoise_step_ms": { - "0": 104.11, - "1": 102.97, - "2": 103.27, - "3": 98.55, - "4": 95.34, - "5": 94.41, - "6": 96.22, - "7": 100.53, - "8": 107.55, - "9": 102.17, - "10": 104.54, - "11": 104.35, - "12": 106.04, - "13": 102.32, - "14": 105.06, - "15": 99.9, - "16": 106.02, - "17": 102.31, - "18": 106.51, - "19": 105.24, - "20": 103.11, - "21": 103.45, - "22": 103.24, - "23": 94.26, - "24": 103.4, - "25": 102.99, - "26": 103.3, - "27": 103.5, - "28": 107.28, - "29": 104.12, - "30": 103.06, - "31": 102.48, - "32": 105.98, - "33": 103.06, - "34": 106.72, - "35": 102.02, - "36": 107.2, - "37": 100.72, - "38": 104.86, - "39": 131.72, - "40": 117.41, - "41": 110.43, - "42": 129.11, - "43": 122.35, - "44": 124.0, - "45": 122.55, - "46": 115.81, - "47": 102.49, - "48": 103.97, - "49": 99.08 + "0": 73.63, + "1": 88.19, + "2": 100.47, + "3": 99.12, + "4": 99.99, + "5": 100.15, + "6": 100.02, + "7": 100.4, + "8": 100.26, + "9": 100.84, + "10": 100.96, + "11": 100.72, + "12": 101.28, + "13": 100.61, + "14": 100.61, + "15": 100.87, + "16": 100.38, + "17": 100.29, + "18": 101.28, + "19": 99.24, + "20": 100.75, + "21": 100.98, + "22": 100.47, + "23": 101.18, + "24": 100.19, + "25": 101.35, + "26": 101.01, + "27": 101.26, + "28": 100.98, + "29": 98.85, + "30": 100.95, + "31": 100.97, + "32": 100.29, + "33": 100.59, + "34": 101.09, + "35": 102.19, + "36": 99.26, + "37": 100.97, + "38": 100.88, + "39": 100.53, + "40": 100.84, + "41": 100.76, + "42": 100.87, + "43": 100.82, + "44": 100.79, + "45": 100.73, + "46": 101.13, + "47": 100.73, + "48": 101.11, + "49": 100.09 }, - "expected_e2e_ms": 5613.23, - "expected_avg_denoise_ms": 105.82, - "expected_median_denoise_ms": 103.47, + "expected_e2e_ms": 5242.81, + "expected_avg_denoise_ms": 99.84, + "expected_median_denoise_ms": 100.74, "estimated_full_test_time_s": 125.5 }, "zimage_image_t2i": { @@ -811,68 +811,68 @@ }, "qwen_image_t2i_cache_dit_enabled": { "stages_ms": { - "InputValidationStage": 0.06, + "InputValidationStage": 0.04, "TextEncodingStage": 540.0, - "LatentPreparationStage": 0.24, - "TimestepPreparationStage": 17.53, - "DenoisingStage": 4279.45, - "DecodingStage": 13.56 + "LatentPreparationStage": 0.16, + "TimestepPreparationStage": 14.34, + "DenoisingStage": 4219.4, + "DecodingStage": 8.11 }, "denoise_step_ms": { - "0": 185.42, - "1": 225.92, - "2": 225.34, - "3": 237.41, - "4": 43.0, - "5": 5.36, - "6": 188.6, - "7": 42.8, - "8": 6.26, - "9": 187.79, - "10": 42.82, - "11": 5.81, - "12": 5.66, - "13": 191.14, - "14": 43.39, - "15": 5.77, - "16": 5.4, - "17": 190.51, - "18": 42.93, - "19": 5.41, - "20": 5.29, - "21": 188.42, - "22": 42.58, - "23": 5.38, - "24": 5.28, - "25": 189.64, - "26": 44.73, - "27": 6.12, - "28": 5.86, - "29": 190.17, - "30": 43.14, - "31": 5.47, - "32": 5.72, - "33": 189.87, - "34": 42.73, - "35": 5.42, - "36": 5.31, - "37": 190.44, - "38": 41.93, - "39": 5.59, - "40": 5.49, - "41": 190.36, - "42": 42.88, - "43": 5.39, - "44": 5.33, - "45": 190.12, - "46": 42.62, - "47": 5.33, - "48": 189.41, - "49": 177.01 + "0": 194.03, + "1": 242.42, + "2": 244.66, + "3": 255.67, + "4": 62.91, + "5": 7.69, + "6": 226.06, + "7": 30.85, + "8": 7.48, + "9": 201.89, + "10": 48.7, + "11": 7.51, + "12": 7.34, + "13": 200.01, + "14": 51.4, + "15": 7.51, + "16": 7.4, + "17": 205.12, + "18": 58.3, + "19": 7.53, + "20": 7.12, + "21": 195.49, + "22": 56.76, + "23": 7.37, + "24": 7.28, + "25": 195.54, + "26": 56.73, + "27": 7.53, + "28": 6.92, + "29": 196.33, + "30": 57.49, + "31": 7.33, + "32": 7.34, + "33": 201.88, + "34": 57.04, + "35": 7.23, + "36": 7.47, + "37": 198.44, + "38": 55.55, + "39": 7.07, + "40": 7.16, + "41": 196.03, + "42": 56.32, + "43": 6.54, + "44": 6.79, + "45": 196.52, + "46": 58.6, + "47": 7.44, + "48": 212.42, + "49": 45.11 }, "expected_e2e_ms": 4800.0, - "expected_avg_denoise_ms": 85.38, - "expected_median_denoise_ms": 57.6, + "expected_avg_denoise_ms": 84.23, + "expected_median_denoise_ms": 53.48, "estimated_full_test_time_s": 124.9 }, "wan2_1_t2v_1.3b_teacache_enabled": { @@ -1096,69 +1096,69 @@ }, "ltx_2_two_stage_t2v": { "stages_ms": { - "InputValidationStage": 0.03, - "TextEncodingStage": 1463.27, - "LTX2TextConnectorStage": 12.85, - "LTX2HalveResolutionStage": 0.11, - "LTX2LoRASwitchStage": 325.46, - "LTX2SigmaPreparationStage": 0.25, - "TimestepPreparationStage": 7.85, - "LTX2AVLatentPreparationStage": 0.34, + "InputValidationStage": 0.04, + "TextEncodingStage": 1570.42, + "LTX2TextConnectorStage": 10.72, + "LTX2HalveResolutionStage": 0.05, + "LTX2LoRASwitchStage": 34.59, + "LTX2SigmaPreparationStage": 0.17, + "TimestepPreparationStage": 11.39, + "LTX2AVLatentPreparationStage": 0.18, "LTX2ImageEncodingStage": 0.02, - "LTX2AVDenoisingStage": 6000.0, - "LTX2UpsampleStage": 2.98, - "LTX2RefinementStage": 666.08, - "LTX2AVDecodingStage": 338.87, + "LTX2AVDenoisingStage": 4708.8, + "LTX2UpsampleStage": 2.52, + "LTX2RefinementStage": 585.98, + "LTX2AVDecodingStage": 206.92, "per_frame_generation": null }, "denoise_step_ms": { - "0": 165.1, - "1": 309.19, - "2": 166.63, - "3": 175.53, - "4": 158.77, - "5": 191.1, - "6": 203.59, - "7": 202.98, - "8": 205.09, - "9": 195.92, - "10": 226.96, - "11": 203.85, - "12": 190.75, - "13": 192.07, - "14": 193.85, - "15": 191.34, - "16": 193.56, - "17": 192.05, - "18": 189.53, - "19": 191.61, - "20": 187.52, - "21": 192.57, - "22": 190.67, - "23": 189.47, - "24": 187.38, - "25": 190.22, - "26": 196.7, - "27": 185.05, - "28": 189.59, - "29": 209.85, - "30": 194.47, - "31": 189.43, - "32": 189.58, - "33": 188.41, - "34": 198.11, - "35": 188.45, - "36": 187.06, - "37": 188.65, - "38": 200.22, - "39": 156.46, - "40": 220.06, - "41": 223.51, - "42": 221.07 + "0": 116.82, + "1": 166.1, + "2": 121.53, + "3": 118.69, + "4": 121.32, + "5": 116.1, + "6": 116.65, + "7": 115.37, + "8": 113.66, + "9": 116.34, + "10": 116.37, + "11": 118.34, + "12": 114.94, + "13": 120.27, + "14": 117.81, + "15": 118.44, + "16": 118.83, + "17": 115.56, + "18": 115.91, + "19": 114.51, + "20": 116.32, + "21": 118.88, + "22": 115.94, + "23": 114.69, + "24": 115.36, + "25": 114.4, + "26": 114.65, + "27": 115.08, + "28": 115.54, + "29": 115.14, + "30": 115.89, + "31": 123.66, + "32": 120.16, + "33": 114.77, + "34": 114.4, + "35": 113.72, + "36": 114.14, + "37": 111.66, + "38": 113.4, + "39": 113.54, + "40": 192.18, + "41": 193.62, + "42": 196.16 }, - "expected_e2e_ms": 8500.0, - "expected_avg_denoise_ms": 150.0, - "expected_median_denoise_ms": 135.0, + "expected_e2e_ms": 7209.55, + "expected_avg_denoise_ms": 122.95, + "expected_median_denoise_ms": 115.94, "estimated_full_test_time_s": 345.4 }, "wan2_2_ti2v_5b": { @@ -1286,85 +1286,85 @@ }, "qwen_image_layered_i2i": { "stages_ms": { - "DecodingStage": 59.13, - "QwenImageLayeredBeforeDenoisingStage": 1724.21, - "DenoisingStage": 33774.04, - "TimestepPreparationStage": 5.02 + "QwenImageLayeredBeforeDenoisingStage": 132.71, + "TimestepPreparationStage": 0.0, + "DenoisingStage": 33400.39, + "DecodingStage": 142.89 }, "denoise_step_ms": { - "0": 520.86, - "1": 684.78, - "2": 674.72, - "3": 673.29, - "4": 679.38, - "5": 674.12, - "6": 677.59, - "7": 677.64, - "8": 673.87, - "9": 677.77, - "10": 675.84, - "11": 677.38, - "12": 675.78, - "13": 673.47, - "14": 678.12, - "15": 676.79, - "16": 677.9, - "17": 679.27, - "18": 678.66, - "19": 677.29, - "20": 679.02, - "21": 676.15, - "22": 678.71, - "23": 676.34, - "24": 677.06, - "25": 677.64, - "26": 678.92, - "27": 681.11, - "28": 679.38, - "29": 678.12, - "30": 679.47, - "31": 680.07, - "32": 680.45, - "33": 675.12, - "34": 678.71, - "35": 680.33, - "36": 676.08, - "37": 677.33, - "38": 679.71, - "39": 678.55, - "40": 675.98, - "41": 675.91, - "42": 676.69, - "43": 675.94, - "44": 678.28, - "45": 675.21, - "46": 676.92, - "47": 674.17, - "48": 676.68, - "49": 676.04 + "0": 512.55, + "1": 675.96, + "2": 663.04, + "3": 666.15, + "4": 669.94, + "5": 668.69, + "6": 663.62, + "7": 672.62, + "8": 670.99, + "9": 672.66, + "10": 673.83, + "11": 671.75, + "12": 670.63, + "13": 672.97, + "14": 672.76, + "15": 675.31, + "16": 672.15, + "17": 671.69, + "18": 672.69, + "19": 670.83, + "20": 670.16, + "21": 671.72, + "22": 665.95, + "23": 670.78, + "24": 668.87, + "25": 672.74, + "26": 668.35, + "27": 672.91, + "28": 669.31, + "29": 669.8, + "30": 673.63, + "31": 666.7, + "32": 670.55, + "33": 666.78, + "34": 670.27, + "35": 675.04, + "36": 668.11, + "37": 671.81, + "38": 671.92, + "39": 669.62, + "40": 674.31, + "41": 672.18, + "42": 670.9, + "43": 669.57, + "44": 674.92, + "45": 673.28, + "46": 674.95, + "47": 673.68, + "48": 674.14, + "49": 670.44 }, - "expected_e2e_ms": 36039.62, - "expected_avg_denoise_ms": 675.35, - "expected_median_denoise_ms": 679.14, + "expected_e2e_ms": 33679.2, + "expected_avg_denoise_ms": 667.88, + "expected_median_denoise_ms": 671.34, "estimated_full_test_time_s": 161.5 }, "fastwan2_2_ti2v_5b": { "stages_ms": { - "InputValidationStage": 380.0, - "TextEncodingStage": 700.0, - "TimestepPreparationStage": 41.12, - "LatentPreparationStage": 0.15, - "DmdDenoisingStage": 428.28, - "DecodingStage": 1550.0 + "InputValidationStage": 400.0, + "TextEncodingStage": 685.05, + "TimestepPreparationStage": 36.07, + "LatentPreparationStage": 0.12, + "DmdDenoisingStage": 401.85, + "DecodingStage": 1524.82 }, "denoise_step_ms": { - "0": 49.59, - "1": 171.65, - "2": 200.43 + "0": 35.0, + "1": 115.0, + "2": 130.0 }, - "expected_e2e_ms": 3150.0, - "expected_avg_denoise_ms": 140.56, - "expected_median_denoise_ms": 171.65, + "expected_e2e_ms": 2684.91, + "expected_avg_denoise_ms": 95.0, + "expected_median_denoise_ms": 120.0, "estimated_full_test_time_s": 125.2 }, "fast_hunyuan_video": { @@ -1389,7 +1389,6 @@ "expected_avg_denoise_ms": 982.57, "expected_median_denoise_ms": 1118.83 }, - "wan2_2_i2v_a14b_2gpu": { "stages_ms": { "InputValidationStage": 18.1, @@ -1763,68 +1762,68 @@ }, "wan2_1_t2v_1_3b_lora_1gpu": { "stages_ms": { - "TimestepPreparationStage": 2.16, - "DenoisingStage": 7527.67, - "DecodingStage": 639.21, - "LatentPreparationStage": 0.14, - "TextEncodingStage": 1073.61, - "InputValidationStage": 0.06 + "InputValidationStage": 0.05, + "TextEncodingStage": 679.6, + "LatentPreparationStage": 0.16, + "TimestepPreparationStage": 3.26, + "DenoisingStage": 7328.48, + "DecodingStage": 563.07 }, "denoise_step_ms": { - "0": 93.39, - "1": 134.18, - "2": 130.54, - "3": 131.78, - "4": 132.36, - "5": 130.92, - "6": 129.94, - "7": 130.08, - "8": 130.84, - "9": 130.55, - "10": 130.16, - "11": 130.12, - "12": 130.85, - "13": 130.17, - "14": 130.1, - "15": 130.22, - "16": 130.04, - "17": 130.45, - "18": 130.23, - "19": 130.33, - "20": 130.08, - "21": 130.12, - "22": 130.21, - "23": 130.44, - "24": 131.08, - "25": 130.35, - "26": 130.27, - "27": 130.33, - "28": 130.22, - "29": 130.45, - "30": 130.32, - "31": 130.09, - "32": 130.74, - "33": 130.44, - "34": 142.71, - "35": 149.8, - "36": 150.24, - "37": 145.42, - "38": 152.33, - "39": 153.65, - "40": 148.9, - "41": 136.97, - "42": 129.88, - "43": 129.91, - "44": 129.76, - "45": 129.52, - "46": 129.79, - "47": 129.66, - "48": 129.58, - "49": 129.55 + "0": 90.0, + "1": 141.31, + "2": 146.15, + "3": 145.23, + "4": 144.57, + "5": 144.41, + "6": 160.18, + "7": 148.95, + "8": 144.97, + "9": 146.95, + "10": 144.53, + "11": 145.21, + "12": 146.25, + "13": 152.87, + "14": 148.99, + "15": 145.88, + "16": 144.81, + "17": 144.76, + "18": 147.29, + "19": 151.36, + "20": 151.05, + "21": 146.67, + "22": 147.05, + "23": 147.91, + "24": 148.1, + "25": 147.86, + "26": 149.47, + "27": 150.5, + "28": 149.59, + "29": 145.76, + "30": 147.28, + "31": 147.92, + "32": 148.82, + "33": 149.74, + "34": 148.97, + "35": 147.51, + "36": 147.15, + "37": 148.49, + "38": 148.15, + "39": 148.62, + "40": 148.49, + "41": 149.09, + "42": 146.27, + "43": 147.88, + "44": 145.54, + "45": 149.11, + "46": 149.56, + "47": 149.23, + "48": 146.76, + "49": 146.49 }, - "expected_e2e_ms": 9624.48, - "expected_avg_denoise_ms": 150.42, - "expected_median_denoise_ms": 151.36, + "expected_e2e_ms": 8579.79, + "expected_avg_denoise_ms": 146.48, + "expected_median_denoise_ms": 147.68, "estimated_full_test_time_s": 129.6 }, "wan2_1_i2v_14b_lora_2gpu": { @@ -2113,68 +2112,68 @@ }, "wan2_1_t2v_1.3b_frame_interp_2x": { "stages_ms": { - "InputValidationStage": 0.06, - "TextEncodingStage": 1121.71, - "LatentPreparationStage": 0.15, - "TimestepPreparationStage": 2.22, - "DenoisingStage": 7541.33, - "DecodingStage": 628.66 + "InputValidationStage": 0.05, + "TextEncodingStage": 677.98, + "LatentPreparationStage": 0.2, + "TimestepPreparationStage": 2.72, + "DenoisingStage": 7336.03, + "DecodingStage": 564.05 }, "denoise_step_ms": { - "0": 81.45, - "1": 154.83, - "2": 151.21, - "3": 150.19, - "4": 150.11, - "5": 157.37, - "6": 154.09, - "7": 152.26, - "8": 151.65, - "9": 149.55, - "10": 150.75, - "11": 154.99, - "12": 152.85, - "13": 151.33, - "14": 150.64, - "15": 149.77, - "16": 151.95, - "17": 154.95, - "18": 152.83, - "19": 152.3, - "20": 151.65, - "21": 150.63, - "22": 151.26, - "23": 153.51, - "24": 152.65, - "25": 152.1, - "26": 151.67, - "27": 151.58, - "28": 152.36, - "29": 153.83, - "30": 151.79, - "31": 151.42, - "32": 151.35, - "33": 151.3, - "34": 153.01, - "35": 152.32, - "36": 152.17, - "37": 151.43, - "38": 152.1, - "39": 151.65, - "40": 153.11, - "41": 152.45, - "42": 152.3, - "43": 151.61, - "44": 151.06, - "45": 151.89, - "46": 152.05, - "47": 152.16, - "48": 152.63, - "49": 150.71 + "0": 88.11, + "1": 143.49, + "2": 146.56, + "3": 145.57, + "4": 145.12, + "5": 144.35, + "6": 158.28, + "7": 148.85, + "8": 144.72, + "9": 144.45, + "10": 145.28, + "11": 144.48, + "12": 150.64, + "13": 154.2, + "14": 147.59, + "15": 146.91, + "16": 148.0, + "17": 144.84, + "18": 147.03, + "19": 151.16, + "20": 150.71, + "21": 146.22, + "22": 146.88, + "23": 145.99, + "24": 145.55, + "25": 148.05, + "26": 149.89, + "27": 149.75, + "28": 148.18, + "29": 147.27, + "30": 147.5, + "31": 145.22, + "32": 149.73, + "33": 151.0, + "34": 146.49, + "35": 147.33, + "36": 148.9, + "37": 145.63, + "38": 147.07, + "39": 149.91, + "40": 150.98, + "41": 146.97, + "42": 148.43, + "43": 148.38, + "44": 145.78, + "45": 149.13, + "46": 149.84, + "47": 150.0, + "48": 146.38, + "49": 147.81 }, - "expected_e2e_ms": 9679.47, - "expected_avg_denoise_ms": 150.71, - "expected_median_denoise_ms": 151.72, + "expected_e2e_ms": 8605.48, + "expected_avg_denoise_ms": 146.61, + "expected_median_denoise_ms": 147.3, "estimated_full_test_time_s": 129.3 }, "flux_2_image_t2i_upscaling_4x": { @@ -2378,50 +2377,51 @@ }, "ltx_2.3_one_stage_ti2v": { "stages_ms": { - "InputValidationStage": 3.05, - "TextEncodingStage": 1677.62, - "LTX2TextConnectorStage": 26.25, + "InputValidationStage": 3.6, + "TextEncodingStage": 1557.57, + "LTX2TextConnectorStage": 27.92, "LTX2SigmaPreparationStage": 0.26, - "TimestepPreparationStage": 13.63, - "LTX2AVLatentPreparationStage": 0.12, - "LTX2AVDenoisingStage": 18500.0, - "LTX2AVDecodingStage": 383.25 + "TimestepPreparationStage": 13.22, + "LTX2AVLatentPreparationStage": 0.17, + "LTX2ImageEncodingStage": 29.61, + "LTX2AVDenoisingStage": 15447.81, + "LTX2AVDecodingStage": 234.96 }, "denoise_step_ms": { - "0": 784.81, - "1": 779.3, - "2": 781.61, - "3": 776.66, - "4": 776.14, - "5": 825.34, - "6": 855.7, - "7": 847.22, - "8": 830.64, - "9": 812.18, - "10": 785.78, - "11": 762.62, - "12": 760.72, - "13": 762.51, - "14": 760.03, - "15": 773.46, - "16": 770.63, - "17": 759.34, - "18": 764.38, - "19": 787.69, - "20": 779.85, - "21": 803.93, - "22": 781.29, - "23": 780.07, - "24": 790.38, - "25": 782.35, - "26": 827.3, - "27": 815.88, - "28": 815.91, - "29": 803.65 + "0": 513.68, + "1": 497.17, + "2": 506.46, + "3": 510.93, + "4": 521.26, + "5": 601.18, + "6": 559.36, + "7": 514.61, + "8": 511.22, + "9": 507.85, + "10": 512.55, + "11": 500.76, + "12": 508.08, + "13": 574.14, + "14": 499.7, + "15": 499.14, + "16": 511.01, + "17": 503.34, + "18": 510.08, + "19": 527.73, + "20": 513.17, + "21": 504.58, + "22": 501.33, + "23": 498.04, + "24": 500.19, + "25": 505.21, + "26": 516.86, + "27": 505.17, + "28": 508.44, + "29": 499.97 }, - "expected_e2e_ms": 21000.0, - "expected_avg_denoise_ms": 620.0, - "expected_median_denoise_ms": 620.0, + "expected_e2e_ms": 17881.82, + "expected_avg_denoise_ms": 514.77, + "expected_median_denoise_ms": 508.26, "estimated_full_test_time_s": 144.2 }, "ltx_2.3_two_stage_t2v_2gpus": { @@ -2483,101 +2483,101 @@ }, "ltx_2_3_two_stage_ti2v_2gpus": { "stages_ms": { - "InputValidationStage": 3.26, - "TextEncodingStage": 1789.58, - "LTX2TextConnectorStage": 30.1, + "InputValidationStage": 3.21, + "TextEncodingStage": 1554.35, + "LTX2TextConnectorStage": 27.64, "LTX2HalveResolutionStage": 0.05, - "LTX2LoRASwitchStage": 127.56, - "LTX2SigmaPreparationStage": 0.54, - "TimestepPreparationStage": 22.06, - "LTX2AVLatentPreparationStage": 0.16, - "LTX2ImageEncodingStage": 27.81, - "LTX2AVDenoisingStage": 8517.54, - "LTX2UpsampleStage": 2.6, - "LTX2RefinementStage": 412.39, - "LTX2AVDecodingStage": 225.41, + "LTX2LoRASwitchStage": 114.37, + "LTX2SigmaPreparationStage": 0.4, + "TimestepPreparationStage": 23.51, + "LTX2AVLatentPreparationStage": 0.15, + "LTX2ImageEncodingStage": 27.58, + "LTX2AVDenoisingStage": 7506.38, + "LTX2UpsampleStage": 2.31, + "LTX2RefinementStage": 382.77, + "LTX2AVDecodingStage": 250.93, "per_frame_generation": null }, "denoise_step_ms": { - "0": 283.67, - "1": 332.51, - "2": 287.41, - "3": 288.1, - "4": 286.66, - "5": 286.86, - "6": 285.56, - "7": 285.01, - "8": 281.79, - "9": 276.41, - "10": 279.14, - "11": 306.21, - "12": 280.42, - "13": 279.25, - "14": 282.39, - "15": 283.28, - "16": 286.48, - "17": 286.48, - "18": 276.96, - "19": 282.12, - "20": 275.69, - "21": 278.08, - "22": 278.56, - "23": 275.05, - "24": 301.15, - "25": 273.35, - "26": 275.35, - "27": 274.09, - "28": 274.32, - "29": 270.66, - "30": 140.67, - "31": 134.03, - "32": 132.51 + "0": 280.97, + "1": 274.21, + "2": 250.26, + "3": 251.97, + "4": 251.58, + "5": 253.73, + "6": 250.99, + "7": 244.5, + "8": 249.33, + "9": 246.74, + "10": 245.15, + "11": 253.89, + "12": 260.57, + "13": 238.39, + "14": 244.41, + "15": 242.77, + "16": 246.85, + "17": 245.53, + "18": 246.1, + "19": 247.94, + "20": 251.52, + "21": 271.21, + "22": 249.64, + "23": 246.86, + "24": 239.52, + "25": 249.31, + "26": 245.69, + "27": 246.51, + "28": 238.94, + "29": 237.2, + "30": 127.08, + "31": 123.1, + "32": 128.06 }, - "expected_e2e_ms": 19149.77, - "expected_avg_denoise_ms": 270.31, - "expected_median_denoise_ms": 280.42, + "expected_e2e_ms": 17660.77, + "expected_avg_denoise_ms": 238.8, + "expected_median_denoise_ms": 246.85, "estimated_full_test_time_s": 170.0 }, "ltx_2_3_hq_pipeline": { "stages_ms": { - "InputValidationStage": 0.09, - "TextEncodingStage": 900.0, + "InputValidationStage": 0.08, + "TextEncodingStage": 803.8, "LTX2TextConnectorStage": 31.22, - "LTX2HalveResolutionStage": 0.12, - "LTX2LoRASwitchStage": 0.03, - "LTX2SigmaPreparationStage": 0.56, - "TimestepPreparationStage": 24.77, - "LTX2AVLatentPreparationStage": 0.33, - "LTX2ImageEncodingStage": 0.04, - "LTX2AVDenoisingStage": 25873.2, - "LTX2UpsampleStage": 752.07, - "LTX2RefinementStage": 2553.3, - "LTX2AVDecodingStage": 585.0, + "LTX2HalveResolutionStage": 0.06, + "LTX2LoRASwitchStage": 0.01, + "LTX2SigmaPreparationStage": 0.27, + "TimestepPreparationStage": 22.21, + "LTX2AVLatentPreparationStage": 0.18, + "LTX2ImageEncodingStage": 0.02, + "LTX2AVDenoisingStage": 16162.58, + "LTX2UpsampleStage": 3.23, + "LTX2RefinementStage": 1593.51, + "LTX2AVDecodingStage": 650.0, "per_frame_generation": null }, "denoise_step_ms": { - "0": 1512.13, - "1": 1506.38, - "2": 1523.98, - "3": 1533.71, - "4": 1569.51, - "5": 1546.87, - "6": 1545.47, - "7": 1528.27, - "8": 1550.64, - "9": 1548.24, - "10": 1531.65, - "11": 2211.84, - "12": 2306.24, - "13": 2270.41, - "14": 1143.7, - "15": 839.75, - "16": 808.41, - "17": 806.56 + "0": 1565.09, + "1": 1034.27, + "2": 989.12, + "3": 1048.34, + "4": 938.4, + "5": 952.6, + "6": 992.87, + "7": 843.37, + "8": 808.52, + "9": 715.13, + "10": 1100.0, + "11": 717.4, + "12": 790.42, + "13": 761.88, + "14": 375.63, + "15": 453.87, + "16": 448.42, + "17": 451.66 }, - "expected_e2e_ms": 30848.37, - "expected_avg_denoise_ms": 1515.76, - "expected_median_denoise_ms": 1532.68, + "expected_e2e_ms": 22500.0, + "expected_avg_denoise_ms": 900.0, + "expected_median_denoise_ms": 980.0, "estimated_full_test_time_s": 363.2 } }