Public API / surface compatibility held.

Recorded

Oct 14, 2026, 08:14 UTC

Since previous step

539ms

State

surface_ok

{
  "files": [
    "python/sglang/kernels/jit/csrc/elementwise/lean_gemma_rmsnorm.cuh",
    "python/sglang/kernels/jit/csrc/gemm/fp8_blockwise/fp8_blockwise_scaled_mm_sm120.cuh",
    "python/sglang/kernels/jit/csrc/gemm/int4n/int4n_gemm.cuh",
    "python/sglang/kernels/jit/csrc/gemm/marlin/gptq_marlin.cuh",
    "python/sglang/kernels/jit/csrc/gemm/marlin/kernel.h",
    "python/sglang/kernels/jit/csrc/gemm/marlin/marlin_template.h",
    "python/sglang/kernels/jit/include/sgl_kernel/utils.cuh",
    "python/sglang/kernels/ops/attention/fla/gdn_replayssm_spec_decode.py",
    "python/sglang/kernels/ops/attention/fla/layernorm_gated.py",
    "python/sglang/kernels/ops/attention/fused_qk_rmsnorm_rope_gate.py",
    "python/sglang/kernels/ops/elementwise/cast_copy.py",
    "python/sglang/kernels/ops/elementwise/elementwise.py",
    "python/sglang/kernels/ops/embeddings/vocab_parallel_embedding.py",
    "python/sglang/kernels/ops/gemm/fp8_blockwise_gemm.py",
    "python/sglang/kernels/ops/kvcache/kv_indices.py",
    "python/sglang/kernels/ops/layernorm/lean_norm.py",
    "python/sglang/kernels/ops/mamba/causal_conv1d_triton.py",
    "python/sglang/kernels/ops/quantization/gptq_marlin.py",
    "python/sglang/kernels/ops/quantization/int4n_gemm.py",
    "python/sglang/kernels/ops/speculative/attn_args.py",
    "python/sglang/kernels/ops/speculative/cache_locs.py",
    "python/sglang/kernels/ops/speculative/commit_indices.py",
    "python/sglang/kernels/ops/speculative/reject_sampling.py",
    "python/sglang/launch_server.py",
    "python/sglang/srt/arg_groups/spec_xqa_defaults.py",
    "python/sglang/srt/arg_groups/speculative_hook.py",
    "python/sglang/srt/environ.py",
    "python/sglang/srt/layers/attention/flashinfer_backend.py",
    "python/sglang/srt/layers/attention/hybrid_linear_attn_backend.py",
    "python/sglang/srt/layers/attention/linear/gdn_backend.py",
    "python/sglang/srt/layers/attention/nvfp4_verify_attn.py",
    "python/sglang/srt/layers/attention/trtllm_mha_backend.py",
    "python/sglang/srt/layers/l2_prefetch.py",
    "python/sglang/srt/layers/layernorm.py",
    "python/sglang/srt/layers/logits_processor.py",
    "python/sglang/srt/layers/quantization/fp4_kv_cache_quant_method.py",
    "python/sglang/srt/layers/quantization/fp8.py",
    "python/sglang/srt/layers/quantization/fp8_utils.py",
    "python/sglang/srt/layers/quantization/marlin_a_prologue.py",
    "python/sglang/srt/layers/quantization/nvfp4_decode.py",
    "python/sglang/srt/layers/quantization/nvfp4_kv_fused_store.py",
    "python/sglang/srt/layers/skinny_linear.py",
    "python/sglang/srt/layers/vocab_parallel_embedding.py",
    "python/sglang/srt/managers/overlap_utils.py",
    "python/sglang/srt/managers/scheduler.py",
    "python/sglang/srt/managers/scheduler_components/batch_result_processor.py",
    "python/sglang/srt/mem_cache/memory_pool.py",
    "python/sglang/srt/model_executor/forward_batch_info.py",
    "python/sglang/srt/model_executor/runner_backend/full_cuda_graph_backend.py",
    "python/sglang/srt/models/qwen2_moe.py",
    "python/sglang/srt/models/qwen3_5.py",
    "python/sglang/srt/models/qwen3_5_mtp.py",
    "python/sglang/srt/models/qwen3_vl.py",
    "python/sglang/srt/speculative/accept_guard.py",
    "python/sglang/srt/speculative/draft_attn_window.py",
    "python/sglang/srt/speculative/draft_fp8_lm_head.py",
    "python/sglang/srt/speculative/draft_vocab_subset.py",
    "python/sglang/srt/speculative/eagle_draft_cuda_graph_runner.py",
    "python/sglang/srt/speculative/eagle_info.py",
    "python/sglang/srt/speculative/eagle_utils.py",
    "python/sglang/srt/speculative/eagle_worker_common.py",
    "python/sglang/srt/speculative/eagle_worker_v2.py",
    "python/sglang/srt/speculative/fast_softmax.py",
    "python/sglang/srt/speculative/spec_utils.py",
    "python/sglang/srt/speculative/subset_sample.py",
    "python/sglang/srt/utils/host_marks.py"
  ]
}