update config (#379)

configs/seko_talk/seko_talk_02_fp8.json

@@ -14,11 +14,11 @@
     "cpu_offload": false,
     "use_31_block": false,
     "clip_quantized": true,
-    "clip_quant_scheme": "fp8",
+    "clip_quant_scheme": "fp8-sgl",
     "dit_quantized": true,
     "dit_quant_scheme": "fp8-sgl",
     "adapter_quantized": true,
-    "adapter_quant_scheme": "fp8",
+    "adapter_quant_scheme": "fp8-sgl",
     "t5_quantized": true,
-    "t5_quant_scheme": "fp8"
+    "t5_quant_scheme": "fp8-sgl"
 }

configs/seko_talk/seko_talk_04_fp8_dist.json

@@ -20,7 +20,7 @@
     "dit_quantized": true,
     "dit_quant_scheme": "fp8-sgl",
     "adapter_quantized": true,
-    "adapter_quant_scheme": "fp8",
+    "adapter_quant_scheme": "fp8-sgl",
     "t5_quantized": true,
-    "t5_quant_scheme": "fp8"
+    "t5_quant_scheme": "fp8-sgl"
 }

configs/seko_talk/seko_talk_05_offload_fp8_4090.json

@@ -24,7 +24,7 @@
     "audio_encoder_cpu_offload": false,
     "audio_adapter_cpu_offload": false,
     "adapter_quantized": true,
-    "adapter_quant_scheme": "fp8",
+    "adapter_quant_scheme": "fp8-q8f",
     "vae_cpu_offload": false,
     "use_tiling_vae": false,
     "dit_quantized": true,

configs/seko_talk/seko_talk_06_offload_fp8_H100.json

@@ -18,12 +18,12 @@
     "t5_cpu_offload": true,
     "t5_offload_granularity": "model",
     "t5_quantized": true,
-    "t5_quant_scheme": "fp8",
+    "t5_quant_scheme": "fp8-sgl",
     "clip_cpu_offload": false,
     "audio_encoder_cpu_offload": false,
     "audio_adapter_cpu_offload": false,
     "adapter_quantized": true,
-    "adapter_quant_scheme": "fp8",
+    "adapter_quant_scheme": "fp8-sgl",
     "vae_cpu_offload": false,
     "use_tiling_vae": false,
     "dit_quantized": true,

configs/seko_talk/seko_talk_10_fp8_dist_fixed_min_area.json

@@ -20,7 +20,7 @@
     "dit_quantized": true,
     "dit_quant_scheme": "fp8-sgl",
     "adapter_quantized": true,
-    "adapter_quant_scheme": "fp8",
+    "adapter_quant_scheme": "fp8-sgl",
     "t5_quantized": true,
-    "t5_quant_scheme": "fp8"
+    "t5_quant_scheme": "fp8-sgl"
 }

configs/seko_talk/seko_talk_11_fp8_dist_fixed_shape.json

@@ -24,7 +24,7 @@
     "dit_quantized": true,
     "dit_quant_scheme": "fp8-sgl",
     "adapter_quantized": true,
-    "adapter_quant_scheme": "fp8",
+    "adapter_quant_scheme": "fp8-sgl",
     "t5_quantized": true,
-    "t5_quant_scheme": "fp8"
+    "t5_quant_scheme": "fp8-sgl"
 }

configs/seko_talk/seko_talk_12_fp8_dist_fixed_shape_8gpus_1s.json

@@ -25,7 +25,7 @@
     "dit_quantized": true,
     "dit_quant_scheme": "fp8-sgl",
     "adapter_quantized": true,
-    "adapter_quant_scheme": "fp8",
+    "adapter_quant_scheme": "fp8-sgl",
     "t5_quantized": true,
-    "t5_quant_scheme": "fp8"
+    "t5_quant_scheme": "fp8-sgl"
 }

configs/seko_talk/seko_talk_13_fp8_dist_bucket_shape_8gpus_5s_realtime.json

@@ -56,7 +56,7 @@
     "dit_quantized": true,
     "dit_quant_scheme": "fp8-sgl",
     "adapter_quantized": true,
-    "adapter_quant_scheme": "fp8",
+    "adapter_quant_scheme": "fp8-sgl",
     "t5_quantized": true,
-    "t5_quant_scheme": "fp8"
+    "t5_quant_scheme": "fp8-sgl"
 }

configs/seko_talk/seko_talk_14_fp8_dist_bucket_shape_8gpus_1s_realtime.json

@@ -57,7 +57,7 @@
     "dit_quantized": true,
     "dit_quant_scheme": "fp8-sgl",
     "adapter_quantized": true,
-    "adapter_quant_scheme": "fp8",
+    "adapter_quant_scheme": "fp8-sgl",
     "t5_quantized": true,
-    "t5_quant_scheme": "fp8"
+    "t5_quant_scheme": "fp8-sgl"
 }

configs/seko_talk/seko_talk_16_fp8_dist_compile.json

@@ -20,9 +20,9 @@
     "dit_quantized": true,
     "dit_quant_scheme": "fp8-sgl",
     "adapter_quantized": true,
-    "adapter_quant_scheme": "fp8",
+    "adapter_quant_scheme": "fp8-sgl",
     "t5_quantized": true,
-    "t5_quant_scheme": "fp8",
+    "t5_quant_scheme": "fp8-sgl",
     "compile": true,
     "compile_shapes": [
         [

lightx2v/models/input_encoders/hf/wan/t5/model.py

@@ -84,9 +84,9 @@ class T5Attention(nn.Module):
         self.head_dim = dim_attn // num_heads
 
         if quantized:
-            if quant_scheme == "int8":
+            if quant_scheme in ["int8", "int8-vllm"]:
                 linear_cls = VllmQuantLinearInt8
-            elif quant_scheme == "fp8":
+            elif quant_scheme in ["fp8", "fp8-sgl"]:
                 linear_cls = SglQuantLinearFp8
             elif quant_scheme == "int8-torchao":
                 linear_cls = TorchaoQuantLinearInt8
@@ -94,6 +94,8 @@ class T5Attention(nn.Module):
                 linear_cls = Q8FQuantLinearInt8
             elif quant_scheme == "fp8-q8f":
                 linear_cls = Q8FQuantLinearFp8
+            else:
+                NotImplementedError(f"Unsupported T5 quant scheme: {quant_scheme}")
         else:
             linear_cls = nn.Linear
 
@@ -151,9 +153,9 @@ class T5FeedForward(nn.Module):
         self.dim_ffn = dim_ffn
 
         if quantized:
-            if quant_scheme == "int8":
+            if quant_scheme in ["int8", "int8-vllm"]:
                 linear_cls = VllmQuantLinearInt8
-            elif quant_scheme == "fp8":
+            elif quant_scheme in ["fp8", "fp8-sgl"]:
                 linear_cls = SglQuantLinearFp8
             elif quant_scheme == "int8-torchao":
                 linear_cls = TorchaoQuantLinearInt8
@@ -161,6 +163,8 @@ class T5FeedForward(nn.Module):
                 linear_cls = Q8FQuantLinearInt8
             elif quant_scheme == "fp8-q8f":
                 linear_cls = Q8FQuantLinearFp8
+            else:
+                NotImplementedError(f"Unsupported T5 quant scheme: {quant_scheme}")
         else:
             linear_cls = nn.Linear
         # layers

lightx2v/models/input_encoders/hf/wan/xlm_roberta/model.py

@@ -59,9 +59,9 @@ class SelfAttention(nn.Module):
 
         # layers
         if quantized:
-            if quant_scheme == "int8":
+            if quant_scheme in ["int8", "int8-vllm"]:
                 linear_cls = VllmQuantLinearInt8
-            elif quant_scheme == "fp8":
+            elif quant_scheme in ["fp8", "fp8-sgl"]:
                 linear_cls = SglQuantLinearFp8
             elif quant_scheme == "int8-torchao":
                 linear_cls = TorchaoQuantLinearInt8
@@ -69,6 +69,8 @@ class SelfAttention(nn.Module):
                 linear_cls = Q8FQuantLinearInt8
             elif quant_scheme == "fp8-q8f":
                 linear_cls = Q8FQuantLinearFp8
+            else:
+                NotImplementedError(f"Unsupported CLip quant scheme: {quant_scheme}")
         else:
             linear_cls = nn.Linear
 
@@ -137,9 +139,9 @@ class AttentionBlock(nn.Module):
 
         # layers
         if quantized:
-            if quant_scheme == "int8":
+            if quant_scheme in ["int8", "int8-vllm"]:
                 linear_cls = VllmQuantLinearInt8
-            elif quant_scheme == "fp8":
+            elif quant_scheme in ["fp8", "fp8-sgl"]:
                 linear_cls = SglQuantLinearFp8
             elif quant_scheme == "int8-torchao":
                 linear_cls = TorchaoQuantLinearInt8
@@ -147,6 +149,8 @@ class AttentionBlock(nn.Module):
                 linear_cls = Q8FQuantLinearInt8
             elif quant_scheme == "fp8-q8f":
                 linear_cls = Q8FQuantLinearFp8
+            else:
+                NotImplementedError(f"Unsupported T5 quant scheme: {quant_scheme}")
         else:
             linear_cls = nn.Linear
 

lightx2v/models/networks/wan/audio_model.py

@@ -28,9 +28,9 @@ class WanAudioModel(WanModel):
     def _load_adapter_ckpt(self):
         if self.config.get("adapter_model_path", None) is None:
             if self.config.get("adapter_quantized", False):
-                if self.config.get("adapter_quant_scheme", None) in ["fp8", "fp8-q8f"]:
+                if self.config.get("adapter_quant_scheme", None) in ["fp8", "fp8-q8f", "fp8-vllm", "fp8-sgl"]:
                     adapter_model_name = "audio_adapter_model_fp8.safetensors"
-                elif self.config.get("adapter_quant_scheme", None) == "int8":
+                elif self.config.get("adapter_quant_scheme", None) in ["int8", "int8-q8f", "int8-vllm", "int8-sgl"]:
                     adapter_model_name = "audio_adapter_model_int8.safetensors"
                 else:
                     raise ValueError(f"Unsupported quant_scheme: {self.config.get('adapter_quant_scheme', None)}")