[plugin] enable OOT platforms to provide custom quant configs (#25347)

Signed-off-by: Devashish Lal <devcode@fb.com>
Co-authored-by: Devashish Lal <devcode@fb.com>
Co-authored-by: Devashish Lal <laldevashish@gmail.com>

docs_new/docs/hardware-platforms/plugin.mdx

@@ -521,6 +521,11 @@ python -c "from sglang.srt.platforms import current_platform; print(current_plat
       <td><code>raise NotImplementedError</code></td>
       <td>Paged allocator class</td>
     </tr>
+    <tr>
+      <td><code>get_quantization_config(quantization)</code></td>
+      <td><code>raise NotImplementedError</code></td>
+      <td>Return hardware-specific quantization config for the specific quantization scheme, raise an error if not supported or return None to use the default config.</td>
+    </tr>
     <tr>
       <td><code>get_piecewise_backend_cls()</code></td>
       <td><code>raise NotImplementedError</code></td>

python/sglang/srt/layers/quantization/__init__.py

@@ -51,6 +51,7 @@ from sglang.srt.layers.quantization.quark_int4fp8_moe import QuarkInt4Fp8Config
 from sglang.srt.layers.quantization.w4afp8 import W4AFp8Config
 from sglang.srt.layers.quantization.w8a8_fp8 import W8A8Fp8Config
 from sglang.srt.layers.quantization.w8a8_int8 import W8A8Int8Config
+from sglang.srt.platforms import current_platform
 from sglang.srt.utils import (
     cpu_has_amx_support,
     is_cpu,
@@ -151,6 +152,13 @@ def get_quantization_config(quantization: str) -> Type[QuantizationConfig]:
         else:
             return CPU_QUANTIZATION_METHODS[quantization]
 
+    if current_platform.is_out_of_tree():
+        config = current_platform.get_quantization_config(quantization)
+
+        # If the platform has a quantization config, use it else use the default
+        if config is not None:
+            return config
+
     return QUANTIZATION_METHODS[quantization]
 
 

python/sglang/srt/platforms/interface.py

@@ -10,8 +10,15 @@ Out-of-tree platforms register via setuptools entry_points under the
 "sglang.srt.platforms" group and should subclass SRTPlatform.
 """
 
+from __future__ import annotations
+
+from typing import TYPE_CHECKING, Optional, Type
+
 from sglang.srt.platforms.device_mixin import DeviceMixin, PlatformEnum
 
+if TYPE_CHECKING:
+    from sglang.srt.layers.quantization.base_config import QuantizationConfig
+
 # Re-export for convenience
 __all__ = ["SRTPlatform", "PlatformEnum"]
 
@@ -80,6 +87,14 @@ class SRTPlatform(DeviceMixin):
         """Return the piecewise compilation backend class for this platform."""
         raise NotImplementedError
 
+    def get_quantization_config(
+        self, quantization: str
+    ) -> Optional[Type[QuantizationConfig]]:
+        """Return hardware-specific quantization config for the specific
+        quantization scheme, raise an error if not supported or return None
+        to use the default config."""
+        return None
+
     # ------------------------------------------------------------------
     # Capability flags (safe conservative defaults)
     # ------------------------------------------------------------------