Revert "works without flash_attn (thanks @juxtapoz!)"

This reverts commit a1b1f86aa3b6780a4981157b8e2e37b0a1017568.

mochi_preview/dit/joint_model/asymm_models_joint.py

@@ -36,11 +36,6 @@ try:
     FLASH_ATTN_IS_AVAILABLE = True
 except ImportError:
     FLASH_ATTN_IS_AVAILABLE = False
-try:
-    from sageattention import sageattn
-    SAGEATTN_IS_AVAILABLE = True
-except ImportError:
-    SAGEATTN_IS_AVAILABLE = False
 
 COMPILE_FINAL_LAYER = False #os.environ.get("COMPILE_DIT") == "1"
 COMPILE_MMDIT_BLOCK = False #os.environ.get("COMPILE_DIT") == "1"
@@ -60,8 +55,9 @@ class AsymmetricAttention(nn.Module):
         attend_to_padding: bool = False,
         softmax_scale: Optional[float] = None,
         device: Optional[torch.device] = None,
-        attention_mode: str = "sdpa",
-        
+        clip_feat_dim: Optional[int] = None,
+        pooled_caption_mlp_bias: bool = True,
+        use_transformer_engine: bool = False,
     ):
         super().__init__()
         self.dim_x = dim_x
@@ -72,7 +68,6 @@ class AsymmetricAttention(nn.Module):
         self.update_y = update_y
         self.attend_to_padding = attend_to_padding
         self.softmax_scale = softmax_scale
-        self.attention_mode = attention_mode
         if dim_x % num_heads != 0:
             raise ValueError(
                 f"dim_x={dim_x} should be divisible by num_heads={num_heads}"
@@ -165,43 +160,6 @@ class AsymmetricAttention(nn.Module):
         )
 
         return qkv
-    
-    def flash_attention(self, qkv, cu_seqlens, max_seqlen_in_batch, total, local_dim):
-        with torch.autocast("cuda", enabled=False):
-            out: torch.Tensor = flash_attn_varlen_qkvpacked_func(
-                qkv,
-                cu_seqlens=cu_seqlens,
-                max_seqlen=max_seqlen_in_batch,
-                dropout_p=0.0,
-                softmax_scale=self.softmax_scale,
-            )  # (total, local_heads, head_dim)
-            return out.view(total, local_dim)
-        
-    def sdpa_attention(self, qkv):
-        q, k, v = rearrange(qkv, '(b s) t h d -> t b h s d', b=1)
-        with torch.autocast("cuda", enabled=False):
-            out = F.scaled_dot_product_attention(
-                q, 
-                k, 
-                v, 
-                attn_mask=None, 
-                dropout_p=0.0, 
-                is_causal=False
-                )
-            return rearrange(out, 'b h s d -> s (b h d)')
-        
-    def sage_attention(self, qkv):
-        q, k, v = rearrange(qkv, '(b s) t h d -> t b h s d', b=1)
-        with torch.autocast("cuda", enabled=False):
-            out = sageattn(
-                q, 
-                k, 
-                v, 
-                attn_mask=None, 
-                dropout_p=0.0, 
-                is_causal=False
-                )
-            return rearrange(out, 'b h s d -> s (b h d)')
 
     @torch.compiler.disable()
     def run_attention(
@@ -222,13 +180,34 @@ class AsymmetricAttention(nn.Module):
         local_dim = local_heads * self.head_dim
         total = qkv.size(0)
 
-        if self.attention_mode == "flash_attn":
-            out = self.flash_attention(qkv, cu_seqlens, max_seqlen_in_batch, total, local_dim)
-        elif self.attention_mode == "sdpa":
-            out = self.sdpa_attention(qkv)
-        elif self.attention_mode == "sage_attn":
-            out = self.sage_attention(qkv)
-        
+        if FLASH_ATTN_IS_AVAILABLE:
+            with torch.autocast("cuda", enabled=False):
+                out: torch.Tensor = flash_attn_varlen_qkvpacked_func(
+                    qkv,
+                    cu_seqlens=cu_seqlens,
+                    max_seqlen=max_seqlen_in_batch,
+                    dropout_p=0.0,
+                    softmax_scale=self.softmax_scale,
+                )  # (total, local_heads, head_dim)
+                out = out.view(total, local_dim)
+        else:
+            raise NotImplementedError("Flash attention is currently required.")
+            print("qkv: ",qkv.shape, qkv.dtype, qkv.device)
+            expected_size = 2 * 44520 * 3 * 24 * 128
+            actual_size = qkv.numel()
+            print(f"Expected size: {expected_size}, Actual size: {actual_size}")
+            q, k, v = qkv.reshape(B, N, 3, local_heads, self.head_dim).permute(2, 0, 3, 1, 4)
+            with torch.autocast("cuda", enabled=False):
+                out = F.scaled_dot_product_attention(
+                    q, 
+                    k, 
+                    v, 
+                    attn_mask=None, 
+                    dropout_p=0.0, 
+                    is_causal=False
+                    )
+                out = out.transpose(1, 2).reshape(B, -1, local_heads * self.head_dim)
+
         x, y = pad_and_split_xy(out, valid_token_indices, B, N, L, qkv.dtype)
         assert x.size() == (B, N, local_dim)
         assert y.size() == (B, L, local_dim)
@@ -305,20 +284,18 @@ class AsymmetricJointBlock(nn.Module):
         mlp_ratio_y: float = 4.0,  # Ratio of hidden size to d_model for MLP for text tokens.
         update_y: bool = True,  # Whether to update text tokens in this block.
         device: Optional[torch.device] = None,
-        attention_mode: str = "sdpa",
         **block_kwargs,
     ):
         super().__init__()
         self.update_y = update_y
         self.hidden_size_x = hidden_size_x
         self.hidden_size_y = hidden_size_y
-        self.attention_mode = attention_mode
         self.mod_x = nn.Linear(hidden_size_x, 4 * hidden_size_x, device=device)
         if self.update_y:
             self.mod_y = nn.Linear(hidden_size_x, 4 * hidden_size_y, device=device)
         else:
             self.mod_y = nn.Linear(hidden_size_x, hidden_size_y, device=device)
-        
+
         # Self-attention:
         self.attn = AsymmetricAttention(
             hidden_size_x,
@@ -326,7 +303,6 @@ class AsymmetricJointBlock(nn.Module):
             num_heads=num_heads,
             update_y=update_y,
             device=device,
-            attention_mode=attention_mode,
             **block_kwargs,
         )
 
@@ -473,7 +449,6 @@ class AsymmDiTJoint(nn.Module):
         use_extended_posenc: bool = False,
         rope_theta: float = 10000.0,
         device: Optional[torch.device] = None,
-        attention_mode: str = "sdpa",
         **block_kwargs,
     ):
         super().__init__()
@@ -541,7 +516,6 @@ class AsymmDiTJoint(nn.Module):
                 mlp_ratio_y=mlp_ratio_y,
                 update_y=update_y,
                 device=device,
-                attention_mode=attention_mode,
                 **block_kwargs,
             )
 

mochi_preview/t2v_synth_mochi.py

@@ -119,7 +119,6 @@ class T2VSynthMochiModel:
         dit_checkpoint_path: str,
         weight_dtype: torch.dtype = torch.float8_e4m3fn,
         fp8_fastmode: bool = False,
-        attention_mode: str = "sdpa"
     ):
         super().__init__()
         self.device = device
@@ -145,7 +144,6 @@ class T2VSynthMochiModel:
                 t5_feat_dim=4096,
                 t5_token_length=256,
                 rope_theta=10000.0,
-                attention_mode=attention_mode,
             )
 
         params_to_keep = {"t_embedder", "x_embedder", "pos_frequencies", "t5", "norm"}

nodes.py

@@ -60,8 +60,7 @@ class DownloadAndLoadMochiModel:
                     {"tooltip": "Downloads from 'https://huggingface.co/Kijai/Mochi_preview_comfy' to 'models/vae/mochi'", },
                 ),
                  "precision": (["fp8_e4m3fn","fp8_e4m3fn_fast","fp16", "fp32", "bf16"],
-                    {"default": "fp8_e4m3fn", }),
-                "attention_mode": (["sdpa","flash_attn","sage_attn"],
+                    {"default": "fp8_e4m3fn", }
                 ),
             },
         }
@@ -72,7 +71,7 @@ class DownloadAndLoadMochiModel:
     CATEGORY = "MochiWrapper"
     DESCRIPTION = "Downloads and loads the selected Mochi model from Huggingface"
 
-    def loadmodel(self, model, vae, precision, attention_mode):
+    def loadmodel(self, model, vae, precision):
 
         device = mm.get_torch_device()
         offload_device = mm.unet_offload_device()
@@ -116,7 +115,6 @@ class DownloadAndLoadMochiModel:
             dit_checkpoint_path=model_path,
             weight_dtype=dtype,
             fp8_fastmode = True if precision == "fp8_e4m3fn_fast" else False,
-            attention_mode=attention_mode
         )
         with (init_empty_weights() if is_accelerate_available else nullcontext()):
             vae = Decoder(