sgl-project/sglang · error · RuntimeError

dcp_kv_mask is not supported for FP4 KV cache.

Error message

dcp_kv_mask is not supported for FP4 KV cache.

What it means

Error "dcp_kv_mask is not supported for FP4 KV cache." thrown in sgl-project/sglang.

Source

Thrown at python/sglang/srt/mem_cache/memory_pool.py:2398

        cache_k: torch.Tensor,
        cache_v: torch.Tensor,
        k_scale: Optional[float] = None,
        v_scale: Optional[float] = None,
        layer_id_override: Optional[int] = None,
        dcp_kv_mask: Optional[torch.Tensor] = None,
    ):
        loc, _, _ = unwrap_write_loc(loc_info)
        # Catch stale slot ids here instead of as illegal-addr / silent KV
        # corruption in the store_kvcache write (gated on SGLANG_ENABLE_ASYNC_ASSERT).
        maybe_detect_oob(loc, 0, self.size + self.page_size, "set_kv_buffer (MHA)")
        layer_id = (
            layer_id_override if layer_id_override is not None else layer.layer_id
        )
        global_layer_id = layer.layer_id if layer is not None else layer_id

        if self.is_quantized_kv_cache:
            if dcp_kv_mask is not None:
                raise RuntimeError("dcp_kv_mask is not supported for FP4 KV cache.")
            self._set_quantized_kv_buffer(
                layer_id,
                global_layer_id,
                loc,
                cache_k,
                cache_v,
                k_scale,
                v_scale,
            )
            return

        if cache_k.dtype != self.dtype:
            if k_scale is not None:
                cache_k.div_(k_scale)
            if v_scale is not None:
                cache_v.div_(v_scale)
            cache_k = cache_k.to(self.dtype)
            cache_v = cache_v.to(self.dtype)

View on GitHub (pinned to 0132848349)

When it happens

Trigger: Thrown at python/sglang/srt/mem_cache/memory_pool.py:2398 when the library encounters an invalid state.

Common situations: See trigger scenarios.


AI-assisted analysis of sgl-project/sglang@0132848349 (2026-08-28). Data as JSON: /api/errors/e828d6be1468e137. Report an issue: GitHub.