[Fix] Fix oom bug (#307)

configs/caching/teacache/wan_ti2v_tea_720p.json

@@ -14,7 +14,6 @@
     "sample_shift": 5.0,
     "enable_cfg": true,
     "cpu_offload": false,
-    "offload_granularity": "model",
     "fps": 24,
     "feature_caching": "Tea",
     "coefficients": [

configs/dist_infer/wan22_moe_i2v_cfg.json

@@ -13,6 +13,8 @@
     "enable_cfg": true,
     "cpu_offload": true,
     "offload_granularity": "model",
+    "t5_cpu_offload": false,
+    "vae_cpu_offload": false,
     "boundary": 0.900,
     "use_image_encoder": false,
     "parallel": {

configs/dist_infer/wan22_moe_i2v_ulysses.json

@@ -13,6 +13,8 @@
     "enable_cfg": true,
     "cpu_offload": true,
     "offload_granularity": "model",
+    "t5_cpu_offload": false,
+    "vae_cpu_offload": false,
     "boundary": 0.900,
     "use_image_encoder": false,
     "parallel": {

configs/dist_infer/wan22_moe_t2v_cfg.json

@@ -13,6 +13,8 @@
     "enable_cfg": true,
     "cpu_offload": true,
     "offload_granularity": "model",
+    "t5_cpu_offload": false,
+    "vae_cpu_offload": false,
     "boundary": 0.875,
     "parallel": {
         "cfg_p_size": 2

configs/dist_infer/wan22_moe_t2v_cfg_ulysses.json

@@ -13,6 +13,8 @@
     "enable_cfg": true,
     "cpu_offload": true,
     "offload_granularity": "model",
+    "t5_cpu_offload": false,
+    "vae_cpu_offload": false,
     "boundary": 0.875,
     "parallel": {
         "seq_p_size": 4,

configs/dist_infer/wan22_moe_t2v_ulysses.json

@@ -13,6 +13,8 @@
     "enable_cfg": true,
     "cpu_offload": true,
     "offload_granularity": "model",
+    "t5_cpu_offload": false,
+    "vae_cpu_offload": false,
     "boundary": 0.875,
     "parallel": {
         "seq_p_size": 4,

configs/dist_infer/wan22_ti2v_i2v_cfg.json

@@ -14,7 +14,6 @@
     "sample_shift": 5.0,
     "enable_cfg": true,
     "cpu_offload": false,
-    "offload_granularity": "model",
     "fps": 24,
     "use_image_encoder": false,
     "parallel": {

configs/dist_infer/wan22_ti2v_i2v_cfg_ulysses.json

@@ -14,7 +14,6 @@
     "sample_shift": 5.0,
     "enable_cfg": true,
     "cpu_offload": false,
-    "offload_granularity": "model",
     "fps": 24,
     "use_image_encoder": false,
     "parallel": {

configs/dist_infer/wan22_ti2v_i2v_ulysses.json

@@ -14,7 +14,6 @@
     "sample_shift": 5.0,
     "enable_cfg": true,
     "cpu_offload": false,
-    "offload_granularity": "model",
     "fps": 24,
     "use_image_encoder": false,
     "parallel": {

configs/dist_infer/wan22_ti2v_t2v_cfg.json

@@ -14,7 +14,6 @@
     "sample_shift": 5.0,
     "enable_cfg": true,
     "cpu_offload": false,
-    "offload_granularity": "model",
     "fps": 24,
     "parallel": {
         "cfg_p_size": 2

configs/dist_infer/wan22_ti2v_t2v_cfg_ulysses.json

@@ -14,7 +14,6 @@
     "sample_shift": 5.0,
     "enable_cfg": true,
     "cpu_offload": false,
-    "offload_granularity": "model",
     "fps": 24,
     "parallel": {
         "seq_p_size": 4,

configs/dist_infer/wan22_ti2v_t2v_ulysses.json

@@ -14,7 +14,6 @@
     "sample_shift": 5.0,
     "enable_cfg": true,
     "cpu_offload": false,
-    "offload_granularity": "model",
     "fps": 24,
     "parallel": {
         "seq_p_size": 4,

configs/seko_talk/seko_talk_08_5B_base.json

@@ -20,7 +20,6 @@
     "sample_shift": 5.0,
     "enable_cfg": false,
     "cpu_offload": false,
-    "offload_granularity": "model",
     "fps": 24,
     "use_image_encoder": false,
     "use_31_block": false,

configs/wan22/wan_distill_moe_flf2v.json

@@ -13,19 +13,9 @@
     "enable_cfg": false,
     "cpu_offload": true,
     "offload_granularity": "model",
+    "t5_cpu_offload": false,
+    "vae_cpu_offload": false,
     "use_image_encoder": false,
     "boundary_step_index": 2,
-    "denoising_step_list": [1000, 750, 500, 250],
-    "lora_configs": [
-        {
-            "name": "low_noise_model",
-            "path": "/path/to/low_noise_lora",
-            "strength": 1.0
-        },
-        {
-            "name": "high_noise_model",
-            "path": "/path/to/high_noise_lora",
-            "strength": 1.0
-        }
-    ]
+    "denoising_step_list": [1000, 750, 500, 250]
 }

configs/wan22/wan_distill_moe_flf2v_fp8.json

@@ -13,6 +13,8 @@
     "enable_cfg": false,
     "cpu_offload": true,
     "offload_granularity": "model",
+    "t5_cpu_offload": false,
+    "vae_cpu_offload": false,
     "use_image_encoder": false,
     "boundary_step_index": 2,
     "denoising_step_list": [1000, 750, 500, 250],

configs/wan22/wan_distill_moe_flf2v_int8.json

+{
+    "infer_steps": 4,
+    "target_video_length": 81,
+    "text_len": 512,
+    "target_height": 720,
+    "target_width": 1280,
+    "self_attn_1_type": "sage_attn2",
+    "cross_attn_1_type": "sage_attn2",
+    "cross_attn_2_type": "sage_attn2",
+    "seed": 42,
+    "sample_guide_scale": [3.5, 3.5],
+    "sample_shift": 16,
+    "enable_cfg": false,
+    "cpu_offload": true,
+    "offload_granularity": "model",
+    "t5_cpu_offload": false,
+    "vae_cpu_offload": false,
+    "use_image_encoder": false,
+    "boundary_step_index": 2,
+    "denoising_step_list": [1000, 750, 500, 250],
+    "mm_config": {
+        "mm_type": "W-int8-channel-sym-A-int8-channel-sym-dynamic-Vllm"
+    },
+    "t5_quantized": true,
+    "t5_quant_scheme": "int8"
+}

configs/wan22/wan_distill_moe_flf2v_with_lora.json

+{
+    "infer_steps": 4,
+    "target_video_length": 81,
+    "text_len": 512,
+    "target_height": 720,
+    "target_width": 1280,
+    "self_attn_1_type": "sage_attn2",
+    "cross_attn_1_type": "sage_attn2",
+    "cross_attn_2_type": "sage_attn2",
+    "seed": 42,
+    "sample_guide_scale": [3.5, 3.5],
+    "sample_shift": 16,
+    "enable_cfg": false,
+    "cpu_offload": true,
+    "offload_granularity": "model",
+    "t5_cpu_offload": false,
+    "vae_cpu_offload": false,
+    "use_image_encoder": false,
+    "boundary_step_index": 2,
+    "denoising_step_list": [1000, 750, 500, 250],
+    "lora_configs": [
+        {
+            "name": "low_noise_model",
+            "path": "/path/to/low_noise_lora",
+            "strength": 1.0
+        },
+        {
+            "name": "high_noise_model",
+            "path": "/path/to/high_noise_lora",
+            "strength": 1.0
+        }
+    ]
+}

configs/wan22/wan_moe_flf2v.json

@@ -9,9 +9,11 @@
     "seed": 42,
     "sample_shift": 16,
     "enable_cfg": true,
-    "cpu_offload": false,
+    "cpu_offload": true,
     "sample_guide_scale": [3.5, 3.5],
     "offload_granularity": "model",
+    "t5_cpu_offload": false,
+    "vae_cpu_offload": false,
     "boundary": 0.900,
     "use_image_encoder": false
 }

configs/wan22/wan_moe_i2v_4090.json

+{
+    "infer_steps": 40,
+    "target_video_length": 81,
+    "text_len": 512,
+    "target_height": 720,
+    "target_width": 1280,
+    "self_attn_1_type": "flash_attn3",
+    "cross_attn_1_type": "flash_attn3",
+    "cross_attn_2_type": "flash_attn3",
+    "seed": 42,
+    "sample_guide_scale": [3.5, 3.5],
+    "sample_shift": 5.0,
+    "enable_cfg": true,
+    "cpu_offload": true,
+    "offload_granularity": "phase",
+    "boundary": 0.900,
+    "use_image_encoder": false
+}

configs/wan22/wan_moe_i2v_audio.json

@@ -16,6 +16,8 @@
     "enable_cfg": false,
     "cpu_offload": true,
     "offload_granularity": "model",
+    "t5_cpu_offload": false,
+    "vae_cpu_offload": false,
     "boundary": 0.900,
     "use_image_encoder": false,
     "use_31_block": false,