From 50844d8dfee256e44d2bcabf32871908bdacf30b Mon Sep 17 00:00:00 2001
From: gaclove <peng.gaoc@gmail.com>
Date: Tue, 21 Oct 2025 17:18:59 +0800
Subject: [PATCH] feat: update quantization configuration to use fp8-sgl as
 default schemes; remove unused quantization backend logic for cleaner
 implementation

---
 bridge.py      | 36 ++++++------------------------------
 data_models.py | 10 ++++------
 nodes.py       | 19 ++++++-------------
 3 files changed, 16 insertions(+), 49 deletions(-)

diff --git a/bridge.py b/bridge.py
index da40f9b..c838dd7 100644
--- a/bridge.py
+++ b/bridge.py
@@ -125,6 +125,7 @@ class LightX2VDefaultConfig:
         "coefficients": None,
         "use_ret_steps": False,
         # Quantization
+        "dit_quant_scheme": DEFAULT_QUANTIZATION_SCHEMES["dit"],
         "t5_quant_scheme": DEFAULT_QUANTIZATION_SCHEMES["t5"],
         "clip_quant_scheme": DEFAULT_QUANTIZATION_SCHEMES["clip"],
         "adapter_quant_scheme": DEFAULT_QUANTIZATION_SCHEMES["adapter"],
@@ -372,24 +373,6 @@ class ModularConfigManager:
 
         return updates
 
-    def _get_mm_type(self, dit_scheme: str, quant_backend: str) -> str:
-        if dit_scheme == "bf16":
-            return "Default"
-
-        base_pattern = f"W-{dit_scheme}-channel-sym-A-{dit_scheme}-channel-sym-dynamic"
-
-        if quant_backend == "vllm":
-            return f"{base_pattern}-Vllm"
-        elif quant_backend == "sgl":
-            suffix = "-Sgl-ActVllm" if dit_scheme == "int8" else "-Sgl"
-            return f"{base_pattern}{suffix}"
-        elif quant_backend == "q8f":
-            return f"{base_pattern}-Q8F"
-        elif quant_backend == "torchao":
-            return f"{base_pattern}-Torchao"
-        else:
-            return "Default"
-
     def apply_quantization_config(self, config: Dict[str, Any]) -> Dict[str, Any]:
         """Apply quantization configuration."""
         updates = {}
@@ -399,27 +382,20 @@ class ModularConfigManager:
         t5_scheme = config.get("t5_quant_scheme", defaults["t5"])
         clip_scheme = config.get("clip_quant_scheme", defaults["clip"])
         adapter_scheme = config.get("adapter_quant_scheme", defaults["adapter"])
-        quant_backend = config.get("quant_op", "vllm")
 
         updates.update(
             {
-                "clip_quantized": clip_scheme != defaults["clip"],
+                "clip_quantized": clip_scheme != "none",
                 "clip_quant_scheme": clip_scheme,
+                "t5_quantized": t5_scheme != "none",
                 "t5_quant_scheme": t5_scheme,
-                "t5_quantized": t5_scheme != defaults["t5"],
-                "adapter_quantized": adapter_scheme != defaults["adapter"],
+                "dit_quantized": dit_scheme != "none",
+                "dit_quant_scheme": dit_scheme,
+                "adapter_quantized": adapter_scheme != "none",
                 "adapter_quant_scheme": adapter_scheme,
             }
         )
 
-        if updates.get("t5_quantized") and quant_backend == "q8f":
-            updates["t5_quant_scheme"] = f"{t5_scheme}-q8f"
-        if updates.get("clip_quantized") and quant_backend == "q8f":
-            updates["clip_quant_scheme"] = f"{clip_scheme}-q8f"
-
-        mm_type = self._get_mm_type(dit_scheme, quant_backend)
-        updates["mm_config"] = {"mm_type": mm_type}
-
         return updates
 
     def apply_memory_optimization(self, config: Dict[str, Any]) -> Dict[str, Any]:
diff --git a/data_models.py b/data_models.py
index ae16cb6..6bfa767 100644
--- a/data_models.py
+++ b/data_models.py
@@ -95,15 +95,13 @@ class TeaCacheConfig:
 class QuantizationConfig:
     """Quantization configuration."""
 
-    quant_op: str = "none"
-    dit_quant_scheme: str = "bf16"
-    t5_quant_scheme: str = "bf16"
-    clip_quant_scheme: str = "fp16"
-    adapter_quant_scheme: str = "bf16"
+    dit_quant_scheme: str = "fp8-sgl"
+    t5_quant_scheme: str = "fp8-sgl"
+    clip_quant_scheme: str = "fp8-sgl"
+    adapter_quant_scheme: str = "fp8-sgl"
 
     def to_dict(self) -> Dict[str, Any]:
         return {
-            "quant_op": self.quant_op,
             "dit_quant_scheme": self.dit_quant_scheme,
             "t5_quant_scheme": self.t5_quant_scheme,
             "clip_quant_scheme": self.clip_quant_scheme,
diff --git a/nodes.py b/nodes.py
index edd4cbc..088a641 100644
--- a/nodes.py
+++ b/nodes.py
@@ -304,21 +304,14 @@ class LightX2VQuantization:
             if is_available:
                 quant_backends.append(op_name)
 
-        # Always have at least one option
-        if not quant_backends:
-            quant_backends = ["none"]
-
-        supported_quant_schemes = ["bf16", "fp16", "fp8", "int8"]
+        common_schema = ["fp8", "int8"]
+        supported_quant_schemes = ["none"]
+        for schema in common_schema:
+            for backend in quant_backends:
+                supported_quant_schemes.append(f"{schema}-{backend}")
 
         return {
             "required": {
-                "quant_op": (
-                    quant_backends,
-                    {
-                        "default": quant_backends[0],
-                        "tooltip": "Quantization computation backend",
-                    },
-                ),
                 "dit_quant_scheme": (
                     supported_quant_schemes,
                     {
@@ -357,7 +350,7 @@ class LightX2VQuantization:
 
     def create_config(
         self,
-        quant_op,
+        dit_quant_schema,
         dit_quant_scheme,
         t5_quant_scheme,
         clip_quant_scheme,