[Recon] remove mm_config and support load single safetensors file (#375)

configs/seko_talk/L40s/8gpu/seko_talk_fp8.json

@@ -14,9 +14,8 @@
     "use_31_block": false,
     "t5_quantized": true,
     "t5_quant_scheme": "fp8-q8f",
-    "mm_config": {
-        "mm_type": "W-fp8-channel-sym-A-fp8-channel-sym-dynamic-Q8F"
-    },
+    "dit_quantized": true,
+    "dit_quant_scheme": "fp8-q8f",
     "adapter_quantized": true,
     "adapter_quant_scheme": "fp8",
     "cpu_offload": false,

configs/seko_talk/multi_person/02_base_fp8.json

@@ -13,9 +13,8 @@
     "enable_cfg": false,
     "cpu_offload": false,
     "use_31_block": false,
-    "mm_config": {
-        "mm_type": "W-fp8-channel-sym-A-fp8-channel-sym-dynamic-Sgl"
-    },
+    "dit_quantized": true,
+    "dit_quant_scheme": "fp8-sgl",
     "adapter_quantized": true,
     "adapter_quant_scheme": "fp8",
     "t5_quantized": true,

configs/seko_talk/multi_person/04_dist_fp8.json

@@ -13,9 +13,8 @@
     "enable_cfg": false,
     "cpu_offload": false,
     "use_31_block": false,
-    "mm_config": {
-        "mm_type": "W-fp8-channel-sym-A-fp8-channel-sym-dynamic-Sgl"
-    },
+    "dit_quantized": true,
+    "dit_quant_scheme": "fp8-sgl",
     "adapter_quantized": true,
     "adapter_quant_scheme": "fp8",
     "t5_quantized": true,

configs/seko_talk/seko_talk_02_fp8.json

@@ -13,9 +13,10 @@
     "enable_cfg": false,
     "cpu_offload": false,
     "use_31_block": false,
-    "mm_config": {
-        "mm_type": "W-fp8-channel-sym-A-fp8-channel-sym-dynamic-Sgl"
-    },
+    "clip_quantized": true,
+    "clip_quant_scheme": "fp8",
+    "dit_quantized": true,
+    "dit_quant_scheme": "fp8-sgl",
     "adapter_quantized": true,
     "adapter_quant_scheme": "fp8",
     "t5_quantized": true,

configs/seko_talk/seko_talk_04_fp8_dist.json

@@ -17,9 +17,8 @@
         "seq_p_size": 8,
         "seq_p_attn_type": "ulysses"
     },
-    "mm_config": {
-        "mm_type": "W-fp8-channel-sym-A-fp8-channel-sym-dynamic-Sgl"
-    },
+    "dit_quantized": true,
+    "dit_quant_scheme": "fp8-sgl",
     "adapter_quantized": true,
     "adapter_quant_scheme": "fp8",
     "t5_quantized": true,

configs/seko_talk/seko_talk_05_offload_fp8_4090.json

@@ -27,7 +27,6 @@
     "adapter_quant_scheme": "fp8",
     "vae_cpu_offload": false,
     "use_tiling_vae": false,
-    "mm_config": {
-        "mm_type": "W-fp8-channel-sym-A-fp8-channel-sym-dynamic-Q8F"
-    }
+    "dit_quantized": true,
+    "dit_quant_scheme": "fp8-sgl"
 }

configs/seko_talk/seko_talk_06_offload_fp8_H100.json

@@ -26,7 +26,6 @@
     "adapter_quant_scheme": "fp8",
     "vae_cpu_offload": false,
     "use_tiling_vae": false,
-    "mm_config": {
-        "mm_type": "W-fp8-channel-sym-A-fp8-channel-sym-dynamic-Sgl"
-    }
+    "dit_quantized": true,
+    "dit_quant_scheme": "fp8-sgl"
 }

configs/seko_talk/seko_talk_10_fp8_dist_fixed_min_area.json

@@ -17,9 +17,8 @@
         "seq_p_size": 4,
         "seq_p_attn_type": "ulysses"
     },
-    "mm_config": {
-        "mm_type": "W-fp8-channel-sym-A-fp8-channel-sym-dynamic-Sgl"
-    },
+    "dit_quantized": true,
+    "dit_quant_scheme": "fp8-sgl",
     "adapter_quantized": true,
     "adapter_quant_scheme": "fp8",
     "t5_quantized": true,

configs/seko_talk/seko_talk_11_fp8_dist_fixed_shape.json

@@ -21,9 +21,8 @@
         "seq_p_size": 4,
         "seq_p_attn_type": "ulysses"
     },
-    "mm_config": {
-        "mm_type": "W-fp8-channel-sym-A-fp8-channel-sym-dynamic-Sgl"
-    },
+    "dit_quantized": true,
+    "dit_quant_scheme": "fp8-sgl",
     "adapter_quantized": true,
     "adapter_quant_scheme": "fp8",
     "t5_quantized": true,

configs/seko_talk/seko_talk_12_fp8_dist_fixed_shape_8gpus_1s.json

@@ -22,9 +22,8 @@
         "seq_p_size": 8,
         "seq_p_attn_type": "ulysses"
     },
-    "mm_config": {
-        "mm_type": "W-fp8-channel-sym-A-fp8-channel-sym-dynamic-Sgl"
-    },
+    "dit_quantized": true,
+    "dit_quant_scheme": "fp8-sgl",
     "adapter_quantized": true,
     "adapter_quant_scheme": "fp8",
     "t5_quantized": true,

configs/seko_talk/seko_talk_13_fp8_dist_bucket_shape_8gpus_5s_realtime.json

@@ -53,9 +53,8 @@
         "seq_p_size": 8,
         "seq_p_attn_type": "ulysses"
     },
-    "mm_config": {
-        "mm_type": "W-fp8-channel-sym-A-fp8-channel-sym-dynamic-Sgl"
-    },
+    "dit_quantized": true,
+    "dit_quant_scheme": "fp8-sgl",
     "adapter_quantized": true,
     "adapter_quant_scheme": "fp8",
     "t5_quantized": true,

configs/seko_talk/seko_talk_14_fp8_dist_bucket_shape_8gpus_1s_realtime.json

@@ -54,9 +54,8 @@
         "seq_p_size": 8,
         "seq_p_attn_type": "ulysses"
     },
-    "mm_config": {
-        "mm_type": "W-fp8-channel-sym-A-fp8-channel-sym-dynamic-Sgl"
-    },
+    "dit_quantized": true,
+    "dit_quant_scheme": "fp8-sgl",
     "adapter_quantized": true,
     "adapter_quant_scheme": "fp8",
     "t5_quantized": true,

configs/seko_talk/seko_talk_16_fp8_dist_compile.json

@@ -17,9 +17,8 @@
         "seq_p_size": 8,
         "seq_p_attn_type": "ulysses"
     },
-    "mm_config": {
-        "mm_type": "W-fp8-channel-sym-A-fp8-channel-sym-dynamic-Sgl"
-    },
+    "dit_quantized": true,
+    "dit_quant_scheme": "fp8-sgl",
     "adapter_quantized": true,
     "adapter_quant_scheme": "fp8",
     "t5_quantized": true,

configs/wan22/wan_animate_4090.json

@@ -19,9 +19,8 @@
     "refert_num": 1,
     "replace_flag": false,
     "fps": 30,
-    "mm_config": {
-        "mm_type": "W-fp8-channel-sym-A-fp8-channel-sym-dynamic-Q8F"
-    },
+    "dit_quantized": true,
+    "dit_quant_scheme": "fp8-sgl",
     "t5_quantized": true,
     "t5_quant_scheme": "fp8",
     "clip_quantized": true,

configs/wan22/wan_animate_replace_4090.json

@@ -21,9 +21,8 @@
     "refert_num": 1,
     "fps": 30,
     "replace_flag": true,
-    "mm_config": {
-        "mm_type": "W-fp8-channel-sym-A-fp8-channel-sym-dynamic-Q8F"
-    },
+    "dit_quantized": true,
+    "dit_quant_scheme": "fp8-q8f",
     "t5_quantized": true,
     "t5_quant_scheme": "fp8",
     "clip_quantized": true,

configs/wan22/wan_distill_moe_flf2v_fp8.json

@@ -25,9 +25,8 @@
         500,
         250
     ],
-    "mm_config": {
-        "mm_type": "W-fp8-channel-sym-A-fp8-channel-sym-dynamic-Sgl"
-    },
+    "dit_quantized": true,
+    "dit_quant_scheme": "fp8-sgl",
     "t5_quantized": true,
     "t5_quant_scheme": "fp8"
 }

configs/wan22/wan_distill_moe_flf2v_int8.json

@@ -25,9 +25,8 @@
         500,
         250
     ],
-    "mm_config": {
-        "mm_type": "W-int8-channel-sym-A-int8-channel-sym-dynamic-Vllm"
-    },
+    "dit_quantized": true,
+    "dit_quant_scheme": "int8-vllm",
     "t5_quantized": true,
     "t5_quant_scheme": "int8"
 }

configs/wan22/wan_moe_i2v_distill.json

@@ -24,7 +24,5 @@
         750,
         500,
         250
-    ],
-    "dit_distill_ckpt_high": "Wan2.2-I2V-A14B/distill_models/high_noise_model/distill_model.safetensors",
-    "dit_distill_ckpt_low": "Wan2.2-I2V-A14B/distill_models/low_noise_model/distill_model.safetensors"
+    ]
 }

configs/wan22/wan_moe_i2v_distill_4090.json

@@ -25,9 +25,8 @@
         500,
         250
     ],
-    "mm_config": {
-        "mm_type": "W-fp8-channel-sym-A-fp8-channel-sym-dynamic-Q8F"
-    },
+    "dit_quantized": true,
+    "dit_quant_scheme": "fp8-q8f",
     "t5_quantized": true,
     "t5_quant_scheme": "fp8-q8f"
 }

configs/wan22/wan_moe_i2v_distill_quant.json

@@ -25,9 +25,8 @@
         500,
         250
     ],
-    "mm_config": {
-        "mm_type": "W-int8-channel-sym-A-int8-channel-sym-dynamic-Vllm"
-    },
+    "dit_quantized": true,
+    "dit_quant_scheme": "fp8-sgl",
     "t5_quantized": true,
-    "t5_quant_scheme": "int8"
+    "t5_quant_scheme": "fp8"
 }