diff --git a/test/registered/models_e2e/test_deepseek_v4_flash_fp4_b200.py b/test/registered/models_e2e/test_deepseek_v4_flash_fp4_b200.py index 122c4ee1a..37e5fc089 100644 --- a/test/registered/models_e2e/test_deepseek_v4_flash_fp4_b200.py +++ b/test/registered/models_e2e/test_deepseek_v4_flash_fp4_b200.py @@ -24,7 +24,8 @@ from sglang.test.test_utils import ( register_cuda_ci(est_time=465, stage="base-c", runner_config="4-gpu-b200") -MODEL = "deepseek-ai/DeepSeek-V4-Flash" +MODEL = "deepseek-ai/DeepSeek-V4-Flash-0731" +MTP_MODEL = "deepseek-ai/DeepSeek-V4-Flash" SERVER_LAUNCH_TIMEOUT = 3600 DEEPEP_CONFIG = '{"normal_dispatch":{"num_sms":96},"normal_combine":{"num_sms":96}}' @@ -33,7 +34,7 @@ _DEEPEP_ENV = { } -class TestDSV4FlashFP4B200( +class TestDSV4FlashFP4B200MTP( SpecDecodingMixin, BasicDecodeCorrectnessMixin, GSM8KMixin, @@ -47,7 +48,7 @@ class TestDSV4FlashFP4B200( @classmethod def setUpClass(cls): - cls.model = try_cached_model(MODEL) + cls.model = try_cached_model(MTP_MODEL) cls.base_url = DEFAULT_URL_FOR_TEST cls.process = popen_launch_server( cls.model, @@ -77,17 +78,17 @@ class TestDSV4FlashFP4B200( kill_process_tree(cls.process.pid) -class TestDSV4FlashFP4B200Balanced( +class TestDSV4FlashFP4B200DSpark( SpecDecodingMixin, BasicDecodeCorrectnessMixin, GSM8KMixin, CustomTestCase, ): - """Balanced recipe: TP=4, DP=4, DeepEP, EAGLE (1-step spec).""" + """LowLatency recipe: TP=4, FP4 (mxfp4), EAGLE spec decoding.""" gsm8k_accuracy_thres = 0.93 - accept_length_thres = 1.8 - bs_1_speed_thres = 100 + accept_length_thres = 4.0 + bs_1_speed_thres = 300 @classmethod def setUpClass(cls): @@ -101,23 +102,10 @@ class TestDSV4FlashFP4B200Balanced( "--trust-remote-code", "--tp", "4", - "--dp", - "4", - "--enable-dp-attention", - "--moe-a2a-backend", - "deepep", "--speculative-algorithm", - "EAGLE", - "--speculative-num-steps", - "1", - "--speculative-eagle-topk", - "1", - "--speculative-num-draft-tokens", - "2", - "--deepep-config", - DEEPEP_CONFIG, + "DSPARK", + "--disable-flashinfer-autotune", ], - env=_DEEPEP_ENV, ) @classmethod @@ -126,7 +114,7 @@ class TestDSV4FlashFP4B200Balanced( kill_process_tree(cls.process.pid) -class TestDSV4FlashFP4NonMTPB200( +class TestDSV4FlashFP4NonSpecB200( BasicDecodeCorrectnessMixin, GSM8KMixin, CustomTestCase ): """Non-MTP recipe: TP=4, DP=4, DeepEP, no speculative decoding.""" diff --git a/test/registered/models_e2e/test_deepseek_v4_flash_fp4_megamoe_b200.py b/test/registered/models_e2e/test_deepseek_v4_flash_fp4_megamoe_b200.py index c1a377a20..4bfd79756 100644 --- a/test/registered/models_e2e/test_deepseek_v4_flash_fp4_megamoe_b200.py +++ b/test/registered/models_e2e/test_deepseek_v4_flash_fp4_megamoe_b200.py @@ -23,7 +23,7 @@ from sglang.test.test_utils import ( register_cuda_ci(est_time=900, stage="extra-b", runner_config="4-gpu-b200") -MODEL = "deepseek-ai/DeepSeek-V4-Flash" +MODEL = "deepseek-ai/DeepSeek-V4-Flash-0731" SERVER_LAUNCH_TIMEOUT = 3600 @@ -64,16 +64,11 @@ class TestDSV4FlashFP4B200W4A8MegaMoE( "--dp", "4", "--enable-dp-attention", + "--enable-dp-lm-head", "--moe-a2a-backend", "megamoe", "--speculative-algorithm", - "EAGLE", - "--speculative-num-steps", - "1", - "--speculative-eagle-topk", - "1", - "--speculative-num-draft-tokens", - "2", + "DSPARK", ], env=_W4A8_MEGAMOE_ENV, ) @@ -111,17 +106,12 @@ class TestDSV4FlashFP4B200W4A4MegaMoE( "--dp", "4", "--enable-dp-attention", + "--enable-dp-lm-head", "--moe-a2a-backend", "megamoe", "--enable-w4a4-mxfp4-megamoe", "--speculative-algorithm", - "EAGLE", - "--speculative-num-steps", - "3", - "--speculative-eagle-topk", - "1", - "--speculative-num-draft-tokens", - "4", + "DSPARK", ], env=_W4A4_MEGAMOE_ENV, )