fix infinitetalk

Author: root

configs/infinitetalk/a800/infinitetalk_mutli_distilled.json

@@ -31,10 +31,10 @@
     "t5_cpu_offload": false,
     "vae_cpu_offload": false,
     "clip_cpu_offload": false,
-    "self_attn_1_type": "sage_attn2",
-    "cross_attn_1_type": "sage_attn2",
-    "cross_attn_2_type": "sage_attn2",
-    "audio_cross_attn_type": "sage_attn2",
+    "self_attn_1_type": "flash_attn2",
+    "cross_attn_1_type": "flash_attn2",
+    "cross_attn_2_type": "flash_attn2",
+    "audio_cross_attn_type": "flash_attn2",
     "text_len": 512,
     "target_height": 720,
     "target_width": 1280,

configs/infinitetalk/a800/infinitetalk_single_distilled.json

@@ -3,16 +3,16 @@
     "target_video_length": 81,
     "motion_frame": 9,
     "target_fps": 25,
-    "video_duration": 300,
+    "video_duration": 60,
     "infinitetalk_mode": "single",
     "infinitetalk_size": "infinitetalk-720",
-    "dit_quantized_ckpt": "/data/nvme5/gushiqiao/models/InfiniteTalk/seko/InfiniteTalk-4StepDistill-Mean-w-ITAudioAdaptorV6.1-int8.safetensors",
+    "dit_quantized_ckpt": "/models/infinitetalk_models_0703/seko/InfiniteTalk-4StepDistill-Mean-w-ITAudioAdaptorV6.1-int8.safetensors",
     "dit_quantized": true,
     "dit_quant_scheme": "int8-sgl",
-    "adapter_model_path": "/data/nvme5/gushiqiao/models/InfiniteTalk/single/single/infinitetalk-int8.safetensors",
+    "adapter_model_path": "/models/infinitetalk_models_0703/single/single/infinitetalk-int8.safetensors",
     "adapter_quantized": true,
     "adapter_quant_scheme": "int8-sgl",
-    "audio_encoder_path": "/data/nvme5/gushiqiao/models/InfiniteTalk/TencentGameMate/chinese-wav2vec2-base",
+    "audio_encoder_path": "/models/infinitetalk_models_0703/TencentGameMate/chinese-wav2vec2-base",
     "clip_quantized": true,
     "clip_quant_scheme": "int8",
     "t5_quantized": true,
@@ -30,10 +30,10 @@
     "t5_cpu_offload": false,
     "vae_cpu_offload": false,
     "clip_cpu_offload": false,
-    "self_attn_1_type": "sage_attn2",
-    "cross_attn_1_type": "sage_attn2",
-    "cross_attn_2_type": "sage_attn2",
-    "audio_cross_attn_type": "sage_attn2",
+    "self_attn_1_type": "flash_attn2",
+    "cross_attn_1_type": "flash_attn2",
+    "cross_attn_2_type": "flash_attn2",
+    "audio_cross_attn_type": "flash_attn2",
     "text_len": 512,
     "target_height": 720,
     "target_width": 1280,
@@ -43,5 +43,9 @@
     "infinitetalk_audio_output_dim": 768,
     "norm_output_audio": true,
     "mxfp8_fuse_enable": false,
-    "use_timestep_transform": true
+    "use_timestep_transform": true,
+    "parallel": {
+        "seq_p_size": 4,
+        "seq_p_attn_type": "ulysses"
+    }
 }

configs/infinitetalk/a800/infinitetalk_single_distilled_8gpus.json

@@ -30,10 +30,10 @@
     "t5_cpu_offload": false,
     "vae_cpu_offload": false,
     "clip_cpu_offload": false,
-    "self_attn_1_type": "sage_attn2",
-    "cross_attn_1_type": "sage_attn2",
-    "cross_attn_2_type": "sage_attn2",
-    "audio_cross_attn_type": "sage_attn2",
+    "self_attn_1_type": "flash_attn2",
+    "cross_attn_1_type": "flash_attn2",
+    "cross_attn_2_type": "flash_attn2",
+    "audio_cross_attn_type": "flash_attn2",
     "text_len": 512,
     "target_height": 720,
     "target_width": 1280,

lightx2v/models/networks/wan/infer/infinitetalk/transformer_infer.py

@@ -134,7 +134,7 @@ def infer_self_attn(self, phase, x, shift_msa, scale_msa, pre_infer_out):
                 use_fp8_comm=self.seq_p_fp8_comm,
                 use_fp4_comm=self.seq_p_fp4_comm,
                 use_tensor_fusion=self.seq_p_tensor_fusion,
-                enable_head_parallel=self.enable_head_parallel,
+                enable_head_parallel=self.seq_p_head_parallel,
                 **attn_running_args,
             )
         else: