Audio/r2v 5b deb (#251)

* [Feat] audio driven for wan2.2-r2v-5b

* refactor: reduce vae args dim &&  timestep init

* update audio pre_infer

* update runner

* update scripts

* [Fix] move ref timestep padding to Scheduler

* fix wan i2v

---------
Co-authored-by: helloyongyang <yongyang1030@163.com>

configs/audio_driven/wan22_ti2v_i2v_audio.json

 {
     "infer_steps": 4,
-    "target_fps": 16,
+    "target_fps": 24,
     "video_duration": 12,
     "audio_sr": 16000,
     "target_video_length": 121,
@@ -20,9 +20,11 @@
     "offload_granularity": "model",
     "fps": 24,
     "use_image_encoder": false,
+    "adaptive_resize": true,
+    "use_31_block": false,
     "lora_configs": [
         {
-        "path": "/data/nvme0/models/wan_ti2v_5b_ref/20250812/model_ema.safetensors",
+        "path": "/mnt/aigc/rtxiang/pretrain/qianhai_weights/lora_model.safetensors",
         "strength": 0.125
         }
     ]

lightx2v/infer.py

@@ -8,7 +8,7 @@ from lightx2v.models.runners.cogvideox.cogvidex_runner import CogvideoxRunner  #
 from lightx2v.models.runners.graph_runner import GraphRunner
 from lightx2v.models.runners.hunyuan.hunyuan_runner import HunyuanRunner  # noqa: F401
 from lightx2v.models.runners.qwen_image.qwen_image_runner import QwenImageRunner  # noqa: F401
-from lightx2v.models.runners.wan.wan_audio_runner import WanAudioRunner  # noqa: F401
+from lightx2v.models.runners.wan.wan_audio_runner import Wan22AudioRunner, WanAudioRunner  # noqa: F401
 from lightx2v.models.runners.wan.wan_causvid_runner import WanCausVidRunner  # noqa: F401
 from lightx2v.models.runners.wan.wan_distill_runner import WanDistillRunner  # noqa: F401
 from lightx2v.models.runners.wan.wan_runner import Wan22MoeRunner, WanRunner  # noqa: F401

lightx2v/models/networks/wan/infer/audio/post_infer.py

@@ -11,6 +11,7 @@ class WanAudioPostInfer(WanPostInfer):
     @torch.compile(disable=not CHECK_ENABLE_GRAPH_MODE())
     def infer(self, x, pre_infer_out):
         x = x[: pre_infer_out.seq_lens[0]]
+        pre_infer_out.grid_sizes[:, 0] -= 1
         x = self.unpatchify(x, pre_infer_out.grid_sizes)
 
         if self.clean_cuda_cache:

lightx2v/models/networks/wan/infer/audio/pre_infer.py

@@ -24,19 +24,17 @@ class WanAudioPreInfer(WanPreInfer):
         self.freq_dim = config["freq_dim"]
         self.dim = config["dim"]
         self.text_len = config["text_len"]
+        self.rope_t_dim = d // 2 - 2 * (d // 6)
         self.clean_cuda_cache = self.config.get("clean_cuda_cache", False)
         self.infer_dtype = GET_DTYPE()
         self.sensitive_layer_dtype = GET_SENSITIVE_DTYPE()
 
-        if config.parallel:
-            self.sp_size = config.parallel.get("seq_p_size", 1)
-        else:
-            self.sp_size = 1
-
     def infer(self, weights, inputs):
         prev_latents = inputs["previmg_encoder_output"]["prev_latents"]
+        hidden_states = self.scheduler.latents
+        if self.config.model_cls != "wan2.2_audio":
             prev_mask = inputs["previmg_encoder_output"]["prev_mask"]
-        hidden_states = torch.cat([self.scheduler.latents, prev_mask, prev_latents], dim=0)
+            hidden_states = torch.cat([hidden_states, prev_mask, prev_latents], dim=0)
 
         x = hidden_states
         t = self.scheduler.timestep_input
@@ -45,11 +43,10 @@ class WanAudioPreInfer(WanPreInfer):
             context = inputs["text_encoder_output"]["context"]
         else:
             context = inputs["text_encoder_output"]["context_null"]
-        # seq_len = self.scheduler.seq_len
 
         clip_fea = inputs["image_encoder_output"]["clip_encoder_out"]
         ref_image_encoder = inputs["image_encoder_output"]["vae_encoder_out"].to(self.scheduler.latents.dtype)
-        # batch_size = len(x)
+
         num_channels, _, height, width = x.shape
         ref_num_channels, ref_num_frames, _, _ = ref_image_encoder.shape
 
@@ -60,7 +57,7 @@ class WanAudioPreInfer(WanPreInfer):
                 device=self.scheduler.latents.device,
             )
             ref_image_encoder = torch.concat([ref_image_encoder, zero_padding], dim=0)
-        y = ref_image_encoder  # 第一个batch维度变成list
+        y = ref_image_encoder
 
         # embeddings
         x = weights.patch_embedding.apply(x.unsqueeze(0))
@@ -70,8 +67,11 @@ class WanAudioPreInfer(WanPreInfer):
 
         y = weights.patch_embedding.apply(y.unsqueeze(0))
         y = y.flatten(2).transpose(1, 2).contiguous()
+        x = torch.cat([x, y], dim=1).squeeze(0)
 
-        x = torch.cat([x, y], dim=1)
+        ####for r2v # zero temporl component corresponding to ref embeddings
+        self.freqs[grid_sizes[0][0] :, : self.rope_t_dim] = 0
+        grid_sizes[:, 0] += 1
 
         embed = sinusoidal_embedding_1d(self.freq_dim, t.flatten())
         if self.sensitive_layer_dtype != self.infer_dtype:
@@ -117,7 +117,7 @@ class WanAudioPreInfer(WanPreInfer):
         return WanPreInferModuleOutput(
             embed=embed,
             grid_sizes=grid_sizes,
-            x=x.squeeze(0),
+            x=x,
             embed0=embed0.squeeze(0),
             seq_lens=seq_lens,
             freqs=self.freqs,

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

@@ -3,7 +3,6 @@ import torch.distributed as dist
 
 from lightx2v.models.input_encoders.hf.seko_audio.audio_adapter import get_q_lens_audio_range
 from lightx2v.models.networks.wan.infer.offload.transformer_infer import WanOffloadTransformerInfer
-from lightx2v.models.networks.wan.infer.utils import compute_freqs_audio, compute_freqs_audio_dist
 
 
 class WanAudioTransformerInfer(WanOffloadTransformerInfer):
@@ -15,21 +14,15 @@ class WanAudioTransformerInfer(WanOffloadTransformerInfer):
     def set_audio_adapter(self, audio_adapter):
         self.audio_adapter = audio_adapter
 
-    @torch.no_grad()
-    def compute_freqs(self, q, grid_sizes, freqs):
-        if self.config["seq_parallel"]:
-            freqs_i = compute_freqs_audio_dist(q.size(0), q.size(2) // 2, grid_sizes, freqs, self.seq_p_group)
-        else:
-            freqs_i = compute_freqs_audio(q.size(2) // 2, grid_sizes, freqs)
-        return freqs_i
-
     @torch.no_grad()
     def post_process(self, x, y, c_gate_msa, pre_infer_out):
         x = super().post_process(x, y, c_gate_msa, pre_infer_out)
 
+        audio_grid_sizes = [row.clone() for row in pre_infer_out.grid_sizes]
+        audio_grid_sizes[0][0] -= 1
         x = self.modify_hidden_states(
             hidden_states=x.to(self.infer_dtype),
-            grid_sizes=pre_infer_out.grid_sizes,
+            grid_sizes=audio_grid_sizes,
             ca_block=self.audio_adapter.ca[self.block_idx],
             audio_encoder_output=pre_infer_out.adapter_output["audio_encoder_output"],
             t_emb=self.scheduler.audio_adapter_t_emb,

lightx2v/models/networks/wan/infer/utils.py

@@ -20,26 +20,6 @@ def compute_freqs(c, grid_sizes, freqs):
     return freqs_i
 
 
-def compute_freqs_audio(c, grid_sizes, freqs):
-    freqs = freqs.split([c - 2 * (c // 3), c // 3, c // 3], dim=1)
-    f, h, w = grid_sizes[0]
-    valid_token_length = f * h * w
-    f = f + 1  ##for r2v add 1 channel
-    seq_len = f * h * w
-    freqs_i = torch.cat(
-        [
-            freqs[0][:f].view(f, 1, 1, -1).expand(f, h, w, -1),  # 时间(帧)编码
-            freqs[1][:h].view(1, h, 1, -1).expand(f, h, w, -1),  # 空间(高度)编码
-            freqs[2][:w].view(1, 1, w, -1).expand(f, h, w, -1),  # 空间(宽度)编码
-        ],
-        dim=-1,
-    ).reshape(seq_len, 1, -1)
-
-    freqs_i[valid_token_length:, :, :f] = 0  ###for r2v # zero temporl component corresponding to ref embeddings
-
-    return freqs_i
-
-
 def compute_freqs_dist(s, c, grid_sizes, freqs, seq_p_group):
     world_size = dist.get_world_size(seq_p_group)
     cur_rank = dist.get_rank(seq_p_group)
@@ -61,31 +41,6 @@ def compute_freqs_dist(s, c, grid_sizes, freqs, seq_p_group):
     return freqs_i_rank
 
 
-def compute_freqs_audio_dist(s, c, grid_sizes, freqs, seq_p_group):
-    world_size = dist.get_world_size(seq_p_group)
-    cur_rank = dist.get_rank(seq_p_group)
-    freqs = freqs.split([c - 2 * (c // 3), c // 3, c // 3], dim=1)
-    f, h, w = grid_sizes[0]
-    valid_token_length = f * h * w
-    f = f + 1
-    seq_len = f * h * w
-    freqs_i = torch.cat(
-        [
-            freqs[0][:f].view(f, 1, 1, -1).expand(f, h, w, -1),
-            freqs[1][:h].view(1, h, 1, -1).expand(f, h, w, -1),
-            freqs[2][:w].view(1, 1, w, -1).expand(f, h, w, -1),
-        ],
-        dim=-1,
-    ).reshape(seq_len, 1, -1)
-
-    freqs_i[valid_token_length:, :, :f] = 0
-
-    freqs_i = pad_freqs(freqs_i, s * world_size)
-    s_per_rank = s
-    freqs_i_rank = freqs_i[(cur_rank * s_per_rank) : ((cur_rank + 1) * s_per_rank), :, :]
-    return freqs_i_rank
-
-
 def compute_freqs_causvid(c, grid_sizes, freqs, start_frame=0):
     freqs = freqs.split([c - 2 * (c // 3), c // 3, c // 3], dim=1)
     f, h, w = grid_sizes[0]

lightx2v/models/runners/wan/wan_audio_runner.py

@@ -21,10 +21,11 @@ from lightx2v.models.networks.wan.audio_model import WanAudioModel
 from lightx2v.models.networks.wan.lora_adapter import WanLoraWrapper
 from lightx2v.models.runners.wan.wan_runner import WanRunner
 from lightx2v.models.schedulers.wan.audio.scheduler import ConsistencyModelScheduler
+from lightx2v.models.video_encoders.hf.wan.vae_2_2 import Wan2_2_VAE
 from lightx2v.utils.envs import *
 from lightx2v.utils.profiler import ProfilingContext, ProfilingContext4Debug
 from lightx2v.utils.registry_factory import RUNNER_REGISTER
-from lightx2v.utils.utils import load_weights, save_to_video, vae_to_comfyui_image
+from lightx2v.utils.utils import find_torch_model_path, load_weights, save_to_video, vae_to_comfyui_image
 
 
 def get_optimal_patched_size_with_sp(patched_h, patched_w, sp_size):
@@ -300,7 +301,8 @@ class WanAudioRunner(WanRunner):  # type:ignore
             self.vae_encoder = self.load_vae_encoder()
 
         img = rearrange(img, "1 C H W -> 1 C 1 H W")
-        vae_encoder_out = self.vae_encoder.encode(img.to(torch.float))[0]
+        vae_encoder_out = self.vae_encoder.encode(img.to(torch.float)).to(GET_DTYPE())
+
         if self.config.get("lazy_load", False) or self.config.get("unload_modules", False):
             del self.vae_encoder
             torch.cuda.empty_cache()
@@ -339,30 +341,32 @@ class WanAudioRunner(WanRunner):  # type:ignore
         if prev_video is not None:
             # Extract and process last frames
             last_frames = prev_video[:, :, -prev_frame_length:].clone().to(device)
+            if self.config.model_cls != "wan2.2_audio":
                 last_frames = self.frame_preprocessor.process_prev_frames(last_frames)
             prev_frames[:, :, :prev_frame_length] = last_frames
+            prev_len = (prev_frame_length - 1) // 4 + 1
+        else:
+            prev_len = 0
 
         if self.config.get("lazy_load", False) or self.config.get("unload_modules", False):
             self.vae_encoder = self.load_vae_encoder()
 
         _, nframe, height, width = self.model.scheduler.latents.shape
         if self.config.model_cls == "wan2.2_audio":
+            if prev_video is not None:
                 prev_latents = self.vae_encoder.encode(prev_frames.to(vae_dtype)).to(dtype)
-            _, prev_mask = self._wan22_masks_like([self.model.scheduler.latents], zero=True, prev_length=prev_latents.shape[1])
             else:
-            prev_latents = self.vae_encoder.encode(prev_frames.to(vae_dtype))[0].to(dtype)
-
-            if prev_video is not None:
-                prev_token_length = (prev_frame_length - 1) // 4 + 1
-                prev_frame_len = max((prev_token_length - 1) * 4 + 1, 0)
+                prev_latents = None
+            prev_mask = self.model.scheduler.mask
         else:
-                prev_frame_len = 0
+            prev_latents = self.vae_encoder.encode(prev_frames.to(vae_dtype)).to(dtype)
 
             frames_n = (nframe - 1) * 4 + 1
             prev_mask = torch.ones((1, frames_n, height, width), device=device, dtype=dtype)
-            prev_mask[:, prev_frame_len:] = 0
+            prev_mask[:, prev_len:] = 0
             prev_mask = self._wan_mask_rearrange(prev_mask)
 
+        if prev_latents is not None:
             if prev_latents.shape[-2:] != (height, width):
                 logger.warning(f"Size mismatch: prev_latents {prev_latents.shape} vs scheduler latents (H={height}, W={width}). Config tgt_h={self.config.tgt_h}, tgt_w={self.config.tgt_w}")
                 prev_latents = torch.nn.functional.interpolate(prev_latents, size=(height, width), mode="bilinear", align_corners=False)
@@ -372,7 +376,7 @@ class WanAudioRunner(WanRunner):  # type:ignore
             torch.cuda.empty_cache()
             gc.collect()
 
-        return {"prev_latents": prev_latents, "prev_mask": prev_mask}
+        return {"prev_latents": prev_latents, "prev_mask": prev_mask, "prev_len": prev_len}
 
     def _wan_mask_rearrange(self, mask: torch.Tensor) -> torch.Tensor:
         """Rearrange mask for WAN model"""
@@ -413,12 +417,11 @@ class WanAudioRunner(WanRunner):  # type:ignore
         audio_features = self.audio_adapter.forward_audio_proj(audio_features, self.model.scheduler.latents.shape[1])
 
         self.inputs["audio_encoder_output"] = audio_features
+        self.inputs["previmg_encoder_output"] = self.prepare_prev_latents(self.prev_video, prev_frame_length=5)
 
         # Reset scheduler for non-first segments
         if segment_idx > 0:
-            self.model.scheduler.reset()
-
-        self.inputs["previmg_encoder_output"] = self.prepare_prev_latents(self.prev_video, prev_frame_length=5)
+            self.model.scheduler.reset(self.inputs["previmg_encoder_output"])
 
     @ProfilingContext4Debug("End run segment")
     def end_run_segment(self):
@@ -600,3 +603,48 @@ class WanAudioRunner(WanRunner):  # type:ignore
 
         ret["target_shape"] = self.config.target_shape
         return ret
+
+
+@RUNNER_REGISTER("wan2.2_audio")
+class Wan22AudioRunner(WanAudioRunner):
+    def __init__(self, config):
+        super().__init__(config)
+
+    def load_vae_decoder(self):
+        # offload config
+        vae_offload = self.config.get("vae_cpu_offload", self.config.get("cpu_offload"))
+        if vae_offload:
+            vae_device = torch.device("cpu")
+        else:
+            vae_device = torch.device("cuda")
+        vae_config = {
+            "vae_pth": find_torch_model_path(self.config, "vae_pth", "Wan2.2_VAE.pth"),
+            "device": vae_device,
+            "cpu_offload": vae_offload,
+            "offload_cache": self.config.get("vae_offload_cache", False),
+        }
+        vae_decoder = Wan2_2_VAE(**vae_config)
+        return vae_decoder
+
+    def load_vae_encoder(self):
+        # offload config
+        vae_offload = self.config.get("vae_cpu_offload", self.config.get("cpu_offload"))
+        if vae_offload:
+            vae_device = torch.device("cpu")
+        else:
+            vae_device = torch.device("cuda")
+        vae_config = {
+            "vae_pth": find_torch_model_path(self.config, "vae_pth", "Wan2.2_VAE.pth"),
+            "device": vae_device,
+            "cpu_offload": vae_offload,
+            "offload_cache": self.config.get("vae_offload_cache", False),
+        }
+        if self.config.task != "i2v":
+            return None
+        else:
+            return Wan2_2_VAE(**vae_config)
+
+    def load_vae(self):
+        vae_encoder = self.load_vae_encoder()
+        vae_decoder = self.load_vae_decoder()
+        return vae_encoder, vae_decoder

lightx2v/models/runners/wan/wan_runner.py

@@ -309,7 +309,7 @@ class WanRunner(DefaultRunner):
                 dim=1,
             ).cuda()
 
-        vae_encoder_out = self.vae_encoder.encode([vae_input])[0]
+        vae_encoder_out = self.vae_encoder.encode(vae_input.unsqueeze(0))
 
         if self.config.get("lazy_load", False) or self.config.get("unload_modules", False):
             del self.vae_encoder

lightx2v/models/schedulers/wan/audio/scheduler.py

 import gc
+import math
 
 import numpy as np
 import torch
@@ -6,12 +7,22 @@ from loguru import logger
 
 from lightx2v.models.schedulers.wan.scheduler import WanScheduler
 from lightx2v.utils.envs import *
+from lightx2v.utils.utils import masks_like
 
 
 class ConsistencyModelScheduler(WanScheduler):
     def __init__(self, config):
         super().__init__(config)
 
+        if self.config.parallel:
+            self.sp_size = self.config.parallel.get("seq_p_size", 1)
+        else:
+            self.sp_size = 1
+
+        if self.config["model_cls"] == "wan2.2_audio":
+            self.prev_latents = None
+            self.prev_len = 0
+
     def set_audio_adapter(self, audio_adapter):
         self.audio_adapter = audio_adapter
 
@@ -23,7 +34,45 @@ class ConsistencyModelScheduler(WanScheduler):
         if self.audio_adapter.cpu_offload:
             self.audio_adapter.time_embedding.to("cpu")
 
-    def prepare(self, image_encoder_output=None):
+        if self.config.model_cls == "wan2.2_audio":
+            _, lat_f, lat_h, lat_w = self.latents.shape
+            F = (lat_f - 1) * self.config.vae_stride[0] + 1
+            per_latent_token_len = lat_h * lat_w // (self.config.patch_size[1] * self.config.patch_size[2])
+            max_seq_len = ((F - 1) // self.config.vae_stride[0] + 1) * per_latent_token_len
+            max_seq_len = int(math.ceil(max_seq_len / self.sp_size)) * self.sp_size
+
+            temp_ts = (self.mask[0][:, ::2, ::2] * self.timestep_input).flatten()
+            self.timestep_input = torch.cat([temp_ts, temp_ts.new_ones(max_seq_len - temp_ts.size(0)) * self.timestep_input]).unsqueeze(0)
+
+            self.timestep_input = torch.cat(
+                [
+                    self.timestep_input,
+                    torch.zeros(
+                        (1, per_latent_token_len),  # padding for reference frame latent
+                        dtype=self.timestep_input.dtype,
+                        device=self.timestep_input.device,
+                    ),
+                ],
+                dim=1,
+            )
+
+    def prepare_latents(self, target_shape, dtype=torch.float32):
+        self.generator = torch.Generator(device=self.device).manual_seed(self.config.seed)
+        self.latents = torch.randn(
+            target_shape[0],
+            target_shape[1],
+            target_shape[2],
+            target_shape[3],
+            dtype=dtype,
+            device=self.device,
+            generator=self.generator,
+        )
+        if self.config["model_cls"] == "wan2.2_audio":
+            self.mask = masks_like(self.latents, zero=True, prev_len=self.prev_len)
+            if self.prev_latents is not None:
+                self.latents = (1.0 - self.mask) * self.prev_latents + self.mask * self.latents
+
+    def prepare(self, previmg_encoder_output=None):
         self.prepare_latents(self.config.target_shape, dtype=torch.float32)
         timesteps = np.linspace(self.num_train_timesteps, 0, self.infer_steps + 1, dtype=np.float32)
 
@@ -43,8 +92,13 @@ class ConsistencyModelScheduler(WanScheduler):
         x0 = sample - model_output * sigma
         x_t_next = x0 * (1 - sigma_next) + sigma_next * torch.randn(x0.shape, dtype=x0.dtype, device=x0.device, generator=self.generator)
         self.latents = x_t_next
+        if self.config["model_cls"] == "wan2.2_audio" and self.prev_latents is not None:
+            self.latents = (1.0 - self.mask) * self.prev_latents + self.mask * self.latents
 
-    def reset(self):
+    def reset(self, previmg_encoder_output=None):
+        if self.config["model_cls"] == "wan2.2_audio":
+            self.prev_latents = previmg_encoder_output["prev_latents"]
+            self.prev_len = previmg_encoder_output["prev_len"]
         self.prepare_latents(self.config.target_shape, dtype=torch.float32)
         gc.collect()
         torch.cuda.empty_cache()

lightx2v/models/video_encoders/hf/wan/vae.py

@@ -892,17 +892,17 @@ class WanVAE:
         self.inv_std = self.inv_std.cuda()
         self.scale = [self.mean, self.inv_std]
 
-    def encode(self, videos):
+    def encode(self, video):
         """
-        videos: A list of videos each with shape [C, T, H, W].
+        video: one video  with shape [1, C, T, H, W].
         """
         if self.cpu_offload:
             self.to_cuda()
 
         if self.use_tiling:
-            out = [self.model.tiled_encode(u.unsqueeze(0).to(self.current_device()), self.scale).float().squeeze(0) for u in videos]
+            out = self.model.tiled_encode(video, self.scale).float().squeeze(0)
         else:
-            out = [self.model.encode(u.unsqueeze(0).to(self.current_device()), self.scale).float().squeeze(0) for u in videos]
+            out = self.model.encode(video, self.scale).float().squeeze(0)
 
         if self.cpu_offload:
             self.to_cpu()

lightx2v/models/video_encoders/hf/wan/vae_2_2.py

@@ -985,10 +985,10 @@ class Wan2_2_VAE:
         self.inv_std = self.inv_std.cuda()
         self.scale = [self.mean, self.inv_std]
 
-    def encode(self, videos):
+    def encode(self, video):
         if self.cpu_offload:
             self.to_cuda()
-        out = self.model.encode(videos.unsqueeze(0), self.scale).float().squeeze(0)
+        out = self.model.encode(video, self.scale).float().squeeze(0)
         if self.cpu_offload:
             self.to_cpu()
         return out

lightx2v/utils/utils.py

@@ -434,17 +434,16 @@ def load_weights(checkpoint_path, cpu_offload=False, remove_key=None):
     return distributed_weight_dict
 
 
-def masks_like(tensor, zero=False, generator=None, p=0.2):
+def masks_like(tensor, zero=False, generator=None, p=0.2, prev_len=1):
     assert isinstance(tensor, torch.Tensor)
     out = torch.ones_like(tensor)
     if zero:
         if generator is not None:
             random_num = torch.rand(1, generator=generator, device=generator.device).item()
             if random_num < p:
-                out[:, 0] = torch.zeros_like(out[:, 0])
+                out[:, :prev_len] = torch.zeros_like(out[:, :prev_len])
         else:
-            out[:, 0] = torch.zeros_like(out[:, 0])
-
+            out[:, :prev_len] = torch.zeros_like(out[:, :prev_len])
     return out
 
 

scripts/wan22/run_wan22_ti2v_i2v_audio.sh

@@ -9,6 +9,8 @@ export CUDA_VISIBLE_DEVICES=0
 # set environment variables
 source ${lightx2v_path}/scripts/base/base.sh
 
+export ENABLE_GRAPH_MODE=false
+export SENSITIVE_LAYER_DTYPE=None
 
 python -m lightx2v.infer \
 --model_cls wan2.2_audio \