[Spec] Remove dead padded_static_len and stale SGLANG_ENABLE_SPEC_V2 references (#30998)

This commit is contained in:
Liangsheng Yin
2026-07-13 15:30:31 -05:00
committed by GitHub
parent 47030b28be
commit e2728ac504
65 changed files with 12 additions and 166 deletions
@@ -125,7 +125,6 @@ class TestDeepSeekR1MXFP4TP4MTPMI35x(unittest.TestCase):
"SGLANG_USE_AITER": "1",
"SGLANG_ENABLE_OVERLAP_PLAN_STREAM": "1",
# Retired on current main, but kept to mirror the reported launch.
"SGLANG_ENABLE_SPEC_V2": "1",
"ROCM_QUICK_REDUCE_QUANTIZATION": "NONE",
"SGLANG_AITER_FP8_PREFILL_ATTN": "1",
"SGLANG_AITER_MLA_PERSIST": "1",
@@ -23,7 +23,6 @@ QWEN3_32B_ENVS = {
"GLOO_SOCKET_IFNAME": "lo",
"HCCL_OP_EXPANSION_MODE": "AIV",
"SGLANG_ENABLE_OVERLAP_PLAN_STREAM": "1",
"SGLANG_ENABLE_SPEC_V2": "1",
"SGLANG_SCHEDULER_DECREASE_PREFILL_IDLE": "1",
"SGLANG_PREFILL_DELAYER_MAX_DELAY_PASSES": "200",
}
@@ -22,7 +22,6 @@ QWEN3_6_27B_64K_PREFIX_ENVS = {
"HCCL_SOCKET_IFNAME": "lo",
"GLOO_SOCKET_IFNAME": "lo",
"HCCL_OP_EXPANSION_MODE": "AIV",
"SGLANG_ENABLE_SPEC_V2": "1",
"SGLANG_ENABLE_OVERLAP_PLAN_STREAM": "1",
"ASCEND_USE_FIA": "1",
"GDN_ATTN_BACKEND_TRITON": "1",
@@ -23,7 +23,6 @@ QWEN3_6_35B_A3B_3K5_1K5_ENVS = {
"GLOO_SOCKET_IFNAME": "lo",
"HCCL_OP_EXPANSION_MODE": "AIV",
"SGLANG_SET_CPU_AFFINITY": "1",
"SGLANG_ENABLE_SPEC_V2": "1",
"SGLANG_ENABLE_OVERLAP_PLAN_STREAM": "0",
"ASCEND_USE_FIA": "1",
"GDN_ATTN_BACKEND_TRITON": "1",
@@ -23,7 +23,6 @@ ENVS = {
"GLOO_SOCKET_IFNAME": "lo",
"HCCL_OP_EXPANSION_MODE": "AIV",
"SGLANG_ENABLE_OVERLAP_PLAN_STREAM": "1",
"SGLANG_ENABLE_SPEC_V2": "1",
"SGLANG_SCHEDULER_DECREASE_PREFILL_IDLE": "1",
"SGLANG_PREFILL_DELAYER_MAX_DELAY_PASSES": "200",
"HCCL_BUFFSIZE": "400",
@@ -44,7 +44,6 @@ GLM_5_1_PD_SEP_DECODE_ENVS = {
"SGLANG_DISAGGREGATION_WAITING_TIMEOUT": "1200",
"SGLANG_SPEC_ENABLE_OVERLAP_REFLOW": "1",
"SGLANG_ENABLE_OVERLAP_PLAN_STREAM": "1",
"SGLANG_ENABLE_SPEC_V2": "1",
"HCCL_BUFFSIZE": "200",
"SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK": "16",
"TASK_QUEUE_ENABLE": "0",
@@ -27,7 +27,6 @@ ENVS = {
"DEEP_NORMAL_MODE_USE_INT8_QUANT": "1",
"SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK": "64",
"HCCL_BUFFSIZE": "4400",
"SGLANG_ENABLE_SPEC_V2": "1",
"SGLANG_ENABLE_OVERLAP_PLAN_STREAM": "1",
"HCCL_SOCKET_IFNAME": NIC_NAME,
"GLOO_SOCKET_IFNAME": NIC_NAME,
@@ -26,7 +26,6 @@ MINIMAX_M2_5_W8A8_4P_IN64K_OUT1K_PREFIX90_ENVS = {
"TASK_QUEUE_ENABLE": "1",
"ASCEND_USE_FIA": "1",
"SGLANG_SET_CPU_AFFINITY": "1",
"SGLANG_ENABLE_SPEC_V2": "1",
"SGLANG_ENABLE_OVERLAP_PLAN_STREAM": "1",
"SGLANG_NPU_FUSED_MOE_MODE": "2",
"SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK": "140000",
@@ -29,7 +29,6 @@ MINIMAX_M2_5_HIGH_THROUGHPUT_ENVS = {
"HCCL_BUFFSIZE": "1024",
"ASCEND_USE_FIA": "1",
"SGLANG_SET_CPU_AFFINITY": "1",
"SGLANG_ENABLE_SPEC_V2": "1",
"SGLANG_ENABLE_OVERLAP_PLAN_STREAM": "1",
"SGLANG_NPU_FUSED_MOE_MODE": "2",
"SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK": "204800",
@@ -26,7 +26,6 @@ QWEN3_8B_ENVS = {
"GLOO_SOCKET_IFNAME": "lo",
"HCCL_OP_EXPANSION_MODE": "AIV",
"SGLANG_ENABLE_OVERLAP_PLAN_STREAM": "1",
"SGLANG_ENABLE_SPEC_V2": "1",
"SGLANG_SCHEDULER_DECREASE_PREFILL_IDLE": "1",
"SGLANG_PREFILL_DELAYER_MAX_DELAY_PASSES": "50",
}
@@ -24,7 +24,6 @@ QWEN3_8B_ENVS = {
"GLOO_SOCKET_IFNAME": "lo",
"HCCL_OP_EXPANSION_MODE": "AIV",
"SGLANG_ENABLE_OVERLAP_PLAN_STREAM": "1",
"SGLANG_ENABLE_SPEC_V2": "1",
}
QWEN3_8B_OTHER_ARGS = [
@@ -27,7 +27,6 @@ QWEN3_30B_A3B_ENVS = {
"GLOO_SOCKET_IFNAME": "lo",
"HCCL_OP_EXPANSION_MODE": "AIV",
"SGLANG_ENABLE_OVERLAP_PLAN_STREAM": "1",
"SGLANG_ENABLE_SPEC_V2": "1",
"SGLANG_SCHEDULER_DECREASE_PREFILL_IDLE": "1",
"SGLANG_PREFILL_DELAYER_MAX_DELAY_PASSES": "200",
"HCCL_BUFFSIZE": "400",
@@ -23,7 +23,6 @@ QWEN3_32B_ENVS = {
"GLOO_SOCKET_IFNAME": "lo",
"HCCL_OP_EXPANSION_MODE": "AIV",
"SGLANG_ENABLE_OVERLAP_PLAN_STREAM": "1",
"SGLANG_ENABLE_SPEC_V2": "1",
"SGLANG_SCHEDULER_DECREASE_PREFILL_IDLE": "1",
"SGLANG_PREFILL_DELAYER_MAX_DELAY_PASSES": "200",
}
@@ -26,7 +26,6 @@ QWEN3_32B_ENVS = {
"GLOO_SOCKET_IFNAME": "lo",
"HCCL_OP_EXPANSION_MODE": "AIV",
"SGLANG_ENABLE_OVERLAP_PLAN_STREAM": "1",
"SGLANG_ENABLE_SPEC_V2": "1",
"SGLANG_SCHEDULER_DECREASE_PREFILL_IDLE": "1",
"SGLANG_PREFILL_DELAYER_MAX_DELAY_PASSES": "100",
"SGLANG_NPU_USE_DEEPGEMM": "1",
@@ -26,7 +26,6 @@ QWEN3_32B_ENVS = {
"GLOO_SOCKET_IFNAME": "lo",
"HCCL_OP_EXPANSION_MODE": "AIV",
"SGLANG_ENABLE_OVERLAP_PLAN_STREAM": "1",
"SGLANG_ENABLE_SPEC_V2": "1",
"SGLANG_SCHEDULER_DECREASE_PREFILL_IDLE": "1",
"SGLANG_PREFILL_DELAYER_MAX_DELAY_PASSES": "100",
"SGLANG_NPU_USE_DEEPGEMM": "1",
@@ -23,7 +23,6 @@ QWEN3_6_27B_1024_ENVS = {
"HCCL_OP_EXPANSION_MODE": "AIV",
"SGLANG_SET_CPU_AFFINITY": "1",
"SGLANG_VIT_ENABLE_CUDA_GRAPH": "1",
"SGLANG_ENABLE_SPEC_V2": "1",
"SGLANG_NPU_PROFILING": "1",
"SGLANG_NPU_PROFILING_STAGE": "prefill",
"SGLANG_ENABLE_OVERLAP_PLAN_STREAM": "1",
@@ -22,7 +22,6 @@ QWEN3_6_27B_1080P_ENVS = {
"HCCL_OP_EXPANSION_MODE": "AIV",
"SGLANG_SET_CPU_AFFINITY": "1",
"SGLANG_VIT_ENABLE_CUDA_GRAPH": "1",
"SGLANG_ENABLE_SPEC_V2": "1",
"SGLANG_NPU_PROFILING": "0",
"SGLANG_NPU_PROFILING_STAGE": "prefill",
"SGLANG_ENABLE_OVERLAP_PLAN_STREAM": "1",
@@ -21,7 +21,6 @@ QWEN3_6_27B_64K_PREFIX_ENVS = {
"HCCL_SOCKET_IFNAME": "lo",
"GLOO_SOCKET_IFNAME": "lo",
"HCCL_OP_EXPANSION_MODE": "AIV",
"SGLANG_ENABLE_SPEC_V2": "1",
"SGLANG_ENABLE_OVERLAP_PLAN_STREAM": "1",
"ASCEND_USE_FIA": "1",
"GDN_ATTN_BACKEND_TRITON": "1",
@@ -25,7 +25,6 @@ QWEN3_6_27B_3K5_1K5_ENVS = {
"GLOO_SOCKET_IFNAME": "lo",
"HCCL_OP_EXPANSION_MODE": "AIV",
"SGLANG_SET_CPU_AFFINITY": "1",
"SGLANG_ENABLE_SPEC_V2": "1",
"SGLANG_ENABLE_OVERLAP_PLAN_STREAM": "0",
"SGLANG_SCHEDULER_DECREASE_PREFILL_IDLE": "1",
"SGLANG_PREFILL_DELAYER_MAX_DELAY_PASSES": "130",
@@ -21,7 +21,6 @@ QWEN3_6_27B_64K_1K_ENVS = {
"GLOO_SOCKET_IFNAME": "lo",
"HCCL_OP_EXPANSION_MODE": "AIV",
"SGLANG_SET_CPU_AFFINITY": "1",
"SGLANG_ENABLE_SPEC_V2": "1",
"SGLANG_ENABLE_OVERLAP_PLAN_STREAM": "1",
}
@@ -22,7 +22,6 @@ QWEN3_6_27B_128K_ENVS = {
"GLOO_SOCKET_IFNAME": "lo",
"HCCL_OP_EXPANSION_MODE": "AIV",
"SGLANG_SET_CPU_AFFINITY": "1",
"SGLANG_ENABLE_SPEC_V2": "1",
"SGLANG_ENABLE_OVERLAP_PLAN_STREAM": "1",
"SGLANG_SCHEDULER_DECREASE_PREFILL_IDLE": "1",
"SGLANG_PREFILL_DELAYER_MAX_DELAY_PASSES": "20",
@@ -21,7 +21,6 @@ QWEN3_6_27B_16K_1k_ENVS = {
"GLOO_SOCKET_IFNAME": "lo",
"HCCL_OP_EXPANSION_MODE": "AIV",
"SGLANG_SET_CPU_AFFINITY": "1",
"SGLANG_ENABLE_SPEC_V2": "1",
"SGLANG_ENABLE_OVERLAP_PLAN_STREAM": "1",
"SGLANG_SCHEDULER_DECREASE_PREFILL_IDLE": "1",
"SGLANG_PREFILL_DELAYER_MAX_DELAY_PASSES": "100",
@@ -22,7 +22,6 @@ QWEN3_6_27B_64K_1K_ENVS = {
"GLOO_SOCKET_IFNAME": "lo",
"HCCL_OP_EXPANSION_MODE": "AIV",
"SGLANG_SET_CPU_AFFINITY": "1",
"SGLANG_ENABLE_SPEC_V2": "1",
"SGLANG_ENABLE_OVERLAP_PLAN_STREAM": "1",
"SGLANG_SCHEDULER_DECREASE_PREFILL_IDLE": "1",
"SGLANG_PREFILL_DELAYER_MAX_DELAY_PASSES": "30",
@@ -23,7 +23,6 @@ QWEN3_6_35B_A3B_128K_1K_ENVS = {
"GLOO_SOCKET_IFNAME": "lo",
"HCCL_OP_EXPANSION_MODE": "AIV",
"SGLANG_SET_CPU_AFFINITY": "1",
"SGLANG_ENABLE_SPEC_V2": "1",
"SGLANG_ENABLE_OVERLAP_PLAN_STREAM": "1",
"ASCEND_USE_FIA": "1",
"SGLANG_PREFILL_DELAYER_MAX_DELAY_PASSES": "20",
@@ -21,7 +21,6 @@ QWEN3_6_35B_A3B_128K_PREFIX_ENVS = {
"GLOO_SOCKET_IFNAME": "lo",
"HCCL_OP_EXPANSION_MODE": "AIV",
"SGLANG_SET_CPU_AFFINITY": "1",
"SGLANG_ENABLE_SPEC_V2": "1",
"SGLANG_ENABLE_OVERLAP_PLAN_STREAM": "1",
"ASCEND_USE_FIA": "1",
"SGLANG_PREFILL_DELAYER_MAX_DELAY_PASSES": "30",
@@ -23,7 +23,6 @@ QWEN3_6_35B_A3B_3K5_1K5_ENVS = {
"GLOO_SOCKET_IFNAME": "lo",
"HCCL_OP_EXPANSION_MODE": "AIV",
"SGLANG_SET_CPU_AFFINITY": "1",
"SGLANG_ENABLE_SPEC_V2": "1",
"SGLANG_ENABLE_OVERLAP_PLAN_STREAM": "0",
"ASCEND_USE_FIA": "1",
}
@@ -22,7 +22,6 @@ QWEN3_6_35B_A3B_64K_1K_ENVS = {
"GLOO_SOCKET_IFNAME": "lo",
"HCCL_OP_EXPANSION_MODE": "AIV",
"SGLANG_SET_CPU_AFFINITY": "1",
"SGLANG_ENABLE_SPEC_V2": "1",
"SGLANG_ENABLE_OVERLAP_PLAN_STREAM": "1",
"ASCEND_USE_FIA": "1",
"SGLANG_PREFILL_DELAYER_MAX_DELAY_PASSES": "1",
@@ -26,7 +26,6 @@ QWEN3_6_35B_A3B_64K_PREFIX_ENVS = {
"GLOO_SOCKET_IFNAME": "lo",
"HCCL_OP_EXPANSION_MODE": "AIV",
"SGLANG_SET_CPU_AFFINITY": "1",
"SGLANG_ENABLE_SPEC_V2": "1",
"SGLANG_ENABLE_OVERLAP_PLAN_STREAM": "0",
"ASCEND_USE_FIA": "1",
"GDN_ATTN_BACKEND_TRITON": "1",
@@ -31,7 +31,6 @@ QWEN3_NEXT_80B_A3B_ENVS = {
"ASCEND_USE_FIA": "1",
"SGLANG_NPU_USE_MULTI_STREAM": "0",
"SGLANG_WARMUP_TIMEOUT": "3600",
"SGLANG_ENABLE_SPEC_V2": "1",
"SGLANG_ENABLE_OVERLAP_PLAN_STREAM": "1",
"FORCE_DRAFT_MODEL_NON_QUANT": "1",
"HCCL_BUFFSIZE": "2000",