FX: Add MPS support

Author: 刘鑫

src/voxcpm/model/voxcpm.py

@@ -85,11 +85,15 @@ def __init__(
         self.patch_size = config.patch_size
         self.device = config.device
         if not torch.cuda.is_available():
-            self.device = "cpu"
+            if torch.backends.mps.is_available():
+                self.device = "mps"
+            else:
+                self.device = "cpu"
+        print(f"Running on device: {self.device}, dtype: {self.config.dtype}")
 
         # Text-Semantic LM
         self.base_lm = MiniCPMModel(config.lm_config)
-        self.base_lm.setup_cache(1, config.max_length, self.device, get_dtype(config.dtype))
+        self.base_lm.setup_cache(1, config.max_length, self.device, get_dtype(self.config.dtype))
 
         self.text_tokenizer = mask_multichar_chinese_tokens(tokenizer)
         self.audio_start_token = 101
@@ -100,7 +104,7 @@ def __init__(
         residual_lm_config.num_hidden_layers = config.residual_lm_num_layers
         residual_lm_config.vocab_size = 0
         self.residual_lm = MiniCPMModel(residual_lm_config)
-        self.residual_lm.setup_cache(1, config.max_length, self.device, get_dtype(config.dtype))
+        self.residual_lm.setup_cache(1, config.max_length, self.device, get_dtype(self.config.dtype))
 
         # Local Encoder
         encoder_config = config.lm_config.model_copy(deep=True)
@@ -132,7 +136,7 @@ def __init__(
             config.lm_config.hidden_size, 
             config.scalar_quantization_latent_dim, 
             config.scalar_quantization_scale
-        ) 
+        )
         self.enc_to_lm_proj = nn.Linear(config.encoder_config.hidden_dim, config.lm_config.hidden_size)
         self.lm_to_dit_proj = nn.Linear(config.lm_config.hidden_size, config.dit_config.hidden_dim)
         self.res_to_dit_proj = nn.Linear(config.lm_config.hidden_size, config.dit_config.hidden_dim)
@@ -271,7 +275,7 @@ def _generate(
 
         text_token = text_token.unsqueeze(0).to(self.device)
         text_mask = text_mask.unsqueeze(0).to(self.device)
-        audio_feat = audio_feat.unsqueeze(0).to(self.device).to(torch.bfloat16)
+        audio_feat = audio_feat.unsqueeze(0).to(self.device).to(get_dtype(self.config.dtype))
         audio_mask = audio_mask.unsqueeze(0).to(self.device)
 
         target_text_length = len(self.text_tokenizer(target_text))
@@ -484,7 +488,7 @@ def _generate_with_prompt_cache(
 
         text_token = text_token.unsqueeze(0).to(self.device)
         text_mask = text_mask.unsqueeze(0).to(self.device)
-        audio_feat = audio_feat.unsqueeze(0).to(self.device).to(torch.bfloat16)
+        audio_feat = audio_feat.unsqueeze(0).to(self.device).to(get_dtype(self.config.dtype))
         audio_mask = audio_mask.unsqueeze(0).to(self.device)
 
         # run inference
@@ -670,7 +674,7 @@ def from_local(cls, path: str, optimize: bool = True):
         )["state_dict"]
 
         model = cls(config, tokenizer, audio_vae)
-        lm_dtype = get_dtype(config.dtype)
+        lm_dtype = get_dtype(model.config.dtype)
         model = model.to(lm_dtype)
         model.audio_vae = model.audio_vae.to(torch.float32)
 

src/voxcpm/modules/minicpm4/model.py

@@ -153,7 +153,12 @@ def forward(
         cos, sin = position_emb
 
         query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin)
-
+        
+        # ref: https://github.com/pytorch/pytorch/issues/163597
+        # there is a bug in MPS for non-contiguous tensors, so we need to make them contiguous
+        query_states = query_states.contiguous()
+        key_states = key_states.contiguous()
+        value_states = value_states.contiguous()
         attn_output = torch.nn.functional.scaled_dot_product_attention(
             query_states,
             key_states,
@@ -198,6 +203,11 @@ def forward_step(
 
         attn_mask = torch.arange(key_cache.size(2), device=key_cache.device) <= position_id
 
+        # ref: https://github.com/pytorch/pytorch/issues/163597
+        # there is a bug in MPS for non-contiguous tensors, so we need to make them contiguous
+        query_states = query_states.contiguous()
+        key_cache = key_cache.contiguous()
+        value_cache = value_cache.contiguous()
         attn_output = torch.nn.functional.scaled_dot_product_attention(
             query_states,
             key_cache,