ComfyUI v0.3.74

comfy/ldm/lumina/model.py

@@ -11,6 +11,7 @@ import comfy.ldm.common_dit
 from comfy.ldm.modules.diffusionmodules.mmdit import TimestepEmbedder
 from comfy.ldm.modules.attention import optimized_attention_masked
 from comfy.ldm.flux.layers import EmbedND
+from comfy.ldm.flux.math import apply_rope
 import comfy.patcher_extension
 
 
@@ -31,6 +32,7 @@ class JointAttention(nn.Module):
         n_heads: int,
         n_kv_heads: Optional[int],
         qk_norm: bool,
+        out_bias: bool = False,
         operation_settings={},
     ):
         """
@@ -59,7 +61,7 @@ class JointAttention(nn.Module):
         self.out = operation_settings.get("operations").Linear(
             n_heads * self.head_dim,
             dim,
-            bias=False,
+            bias=out_bias,
             device=operation_settings.get("device"),
             dtype=operation_settings.get("dtype"),
         )
@@ -70,35 +72,6 @@ class JointAttention(nn.Module):
         else:
             self.q_norm = self.k_norm = nn.Identity()
 
-    @staticmethod
-    def apply_rotary_emb(
-        x_in: torch.Tensor,
-        freqs_cis: torch.Tensor,
-    ) -> torch.Tensor:
-        """
-        Apply rotary embeddings to input tensors using the given frequency
-        tensor.
-
-        This function applies rotary embeddings to the given query 'xq' and
-        key 'xk' tensors using the provided frequency tensor 'freqs_cis'. The
-        input tensors are reshaped as complex numbers, and the frequency tensor
-        is reshaped for broadcasting compatibility. The resulting tensors
-        contain rotary embeddings and are returned as real tensors.
-
-        Args:
-            x_in (torch.Tensor): Query or Key tensor to apply rotary embeddings.
-            freqs_cis (torch.Tensor): Precomputed frequency tensor for complex
-                exponentials.
-
-        Returns:
-            Tuple[torch.Tensor, torch.Tensor]: Tuple of modified query tensor
-                and key tensor with rotary embeddings.
-        """
-
-        t_ = x_in.reshape(*x_in.shape[:-1], -1, 1, 2)
-        t_out = freqs_cis[..., 0] * t_[..., 0] + freqs_cis[..., 1] * t_[..., 1]
-        return t_out.reshape(*x_in.shape)
-
     def forward(
         self,
         x: torch.Tensor,
@@ -134,8 +107,7 @@ class JointAttention(nn.Module):
         xq = self.q_norm(xq)
         xk = self.k_norm(xk)
 
-        xq = JointAttention.apply_rotary_emb(xq, freqs_cis=freqs_cis)
-        xk = JointAttention.apply_rotary_emb(xk, freqs_cis=freqs_cis)
+        xq, xk = apply_rope(xq, xk, freqs_cis)
 
         n_rep = self.n_local_heads // self.n_local_kv_heads
         if n_rep >= 1:
@@ -215,6 +187,8 @@ class JointTransformerBlock(nn.Module):
         norm_eps: float,
         qk_norm: bool,
         modulation=True,
+        z_image_modulation=False,
+        attn_out_bias=False,
         operation_settings={},
     ) -> None:
         """
@@ -235,10 +209,10 @@ class JointTransformerBlock(nn.Module):
         super().__init__()
         self.dim = dim
         self.head_dim = dim // n_heads
-        self.attention = JointAttention(dim, n_heads, n_kv_heads, qk_norm, operation_settings=operation_settings)
+        self.attention = JointAttention(dim, n_heads, n_kv_heads, qk_norm, out_bias=attn_out_bias, operation_settings=operation_settings)
         self.feed_forward = FeedForward(
             dim=dim,
-            hidden_dim=4 * dim,
+            hidden_dim=dim,
             multiple_of=multiple_of,
             ffn_dim_multiplier=ffn_dim_multiplier,
             operation_settings=operation_settings,
@@ -252,16 +226,27 @@ class JointTransformerBlock(nn.Module):
 
         self.modulation = modulation
         if modulation:
-            self.adaLN_modulation = nn.Sequential(
-                nn.SiLU(),
-                operation_settings.get("operations").Linear(
-                    min(dim, 1024),
-                    4 * dim,
-                    bias=True,
-                    device=operation_settings.get("device"),
-                    dtype=operation_settings.get("dtype"),
-                ),
-            )
+            if z_image_modulation:
+                self.adaLN_modulation = nn.Sequential(
+                    operation_settings.get("operations").Linear(
+                        min(dim, 256),
+                        4 * dim,
+                        bias=True,
+                        device=operation_settings.get("device"),
+                        dtype=operation_settings.get("dtype"),
+                    ),
+                )
+            else:
+                self.adaLN_modulation = nn.Sequential(
+                    nn.SiLU(),
+                    operation_settings.get("operations").Linear(
+                        min(dim, 1024),
+                        4 * dim,
+                        bias=True,
+                        device=operation_settings.get("device"),
+                        dtype=operation_settings.get("dtype"),
+                    ),
+                )
 
     def forward(
         self,
@@ -323,7 +308,7 @@ class FinalLayer(nn.Module):
     The final layer of NextDiT.
     """
 
-    def __init__(self, hidden_size, patch_size, out_channels, operation_settings={}):
+    def __init__(self, hidden_size, patch_size, out_channels, z_image_modulation=False, operation_settings={}):
         super().__init__()
         self.norm_final = operation_settings.get("operations").LayerNorm(
             hidden_size,
@@ -340,10 +325,15 @@ class FinalLayer(nn.Module):
             dtype=operation_settings.get("dtype"),
         )
 
+        if z_image_modulation:
+            min_mod = 256
+        else:
+            min_mod = 1024
+
         self.adaLN_modulation = nn.Sequential(
             nn.SiLU(),
             operation_settings.get("operations").Linear(
-                min(hidden_size, 1024),
+                min(hidden_size, min_mod),
                 hidden_size,
                 bias=True,
                 device=operation_settings.get("device"),
@@ -373,12 +363,16 @@ class NextDiT(nn.Module):
         n_heads: int = 32,
         n_kv_heads: Optional[int] = None,
         multiple_of: int = 256,
-        ffn_dim_multiplier: Optional[float] = None,
+        ffn_dim_multiplier: float = 4.0,
         norm_eps: float = 1e-5,
         qk_norm: bool = False,
         cap_feat_dim: int = 5120,
         axes_dims: List[int] = (16, 56, 56),
         axes_lens: List[int] = (1, 512, 512),
+        rope_theta=10000.0,
+        z_image_modulation=False,
+        time_scale=1.0,
+        pad_tokens_multiple=None,
         image_model=None,
         device=None,
         dtype=None,
@@ -390,6 +384,8 @@ class NextDiT(nn.Module):
         self.in_channels = in_channels
         self.out_channels = in_channels
         self.patch_size = patch_size
+        self.time_scale = time_scale
+        self.pad_tokens_multiple = pad_tokens_multiple
 
         self.x_embedder = operation_settings.get("operations").Linear(
             in_features=patch_size * patch_size * in_channels,
@@ -411,6 +407,7 @@ class NextDiT(nn.Module):
                     norm_eps,
                     qk_norm,
                     modulation=True,
+                    z_image_modulation=z_image_modulation,
                     operation_settings=operation_settings,
                 )
                 for layer_id in range(n_refiner_layers)
@@ -434,7 +431,7 @@ class NextDiT(nn.Module):
             ]
         )
 
-        self.t_embedder = TimestepEmbedder(min(dim, 1024), **operation_settings)
+        self.t_embedder = TimestepEmbedder(min(dim, 1024), output_size=256 if z_image_modulation else None, **operation_settings)
         self.cap_embedder = nn.Sequential(
             operation_settings.get("operations").RMSNorm(cap_feat_dim, eps=norm_eps, elementwise_affine=True, device=operation_settings.get("device"), dtype=operation_settings.get("dtype")),
             operation_settings.get("operations").Linear(
@@ -457,18 +454,24 @@ class NextDiT(nn.Module):
                     ffn_dim_multiplier,
                     norm_eps,
                     qk_norm,
+                    z_image_modulation=z_image_modulation,
+                    attn_out_bias=False,
                     operation_settings=operation_settings,
                 )
                 for layer_id in range(n_layers)
             ]
         )
         self.norm_final = operation_settings.get("operations").RMSNorm(dim, eps=norm_eps, elementwise_affine=True, device=operation_settings.get("device"), dtype=operation_settings.get("dtype"))
-        self.final_layer = FinalLayer(dim, patch_size, self.out_channels, operation_settings=operation_settings)
+        self.final_layer = FinalLayer(dim, patch_size, self.out_channels, z_image_modulation=z_image_modulation, operation_settings=operation_settings)
+
+        if self.pad_tokens_multiple is not None:
+            self.x_pad_token = nn.Parameter(torch.empty((1, dim), device=device, dtype=dtype))
+            self.cap_pad_token = nn.Parameter(torch.empty((1, dim), device=device, dtype=dtype))
 
         assert (dim // n_heads) == sum(axes_dims)
         self.axes_dims = axes_dims
         self.axes_lens = axes_lens
-        self.rope_embedder = EmbedND(dim=dim // n_heads, theta=10000.0, axes_dim=axes_dims)
+        self.rope_embedder = EmbedND(dim=dim // n_heads, theta=rope_theta, axes_dim=axes_dims)
         self.dim = dim
         self.n_heads = n_heads
 
@@ -503,108 +506,42 @@ class NextDiT(nn.Module):
         bsz = len(x)
         pH = pW = self.patch_size
         device = x[0].device
-        dtype = x[0].dtype
 
-        if cap_mask is not None:
-            l_effective_cap_len = cap_mask.sum(dim=1).tolist()
-        else:
-            l_effective_cap_len = [num_tokens] * bsz
+        if self.pad_tokens_multiple is not None:
+            pad_extra = (-cap_feats.shape[1]) % self.pad_tokens_multiple
+            cap_feats = torch.cat((cap_feats, self.cap_pad_token.to(device=cap_feats.device, dtype=cap_feats.dtype).unsqueeze(0).repeat(cap_feats.shape[0], pad_extra, 1)), dim=1)
 
-        if cap_mask is not None and not torch.is_floating_point(cap_mask):
-            cap_mask = (cap_mask - 1).to(dtype) * torch.finfo(dtype).max
+        cap_pos_ids = torch.zeros(bsz, cap_feats.shape[1], 3, dtype=torch.float32, device=device)
+        cap_pos_ids[:, :, 0] = torch.arange(cap_feats.shape[1], dtype=torch.float32, device=device) + 1.0
 
-        img_sizes = [(img.size(1), img.size(2)) for img in x]
-        l_effective_img_len = [(H // pH) * (W // pW) for (H, W) in img_sizes]
+        B, C, H, W = x.shape
+        x = self.x_embedder(x.view(B, C, H // pH, pH, W // pW, pW).permute(0, 2, 4, 3, 5, 1).flatten(3).flatten(1, 2))
 
-        max_seq_len = max(
-            (cap_len+img_len for cap_len, img_len in zip(l_effective_cap_len, l_effective_img_len))
-        )
-        max_cap_len = max(l_effective_cap_len)
-        max_img_len = max(l_effective_img_len)
+        H_tokens, W_tokens = H // pH, W // pW
+        x_pos_ids = torch.zeros((bsz, x.shape[1], 3), dtype=torch.float32, device=device)
+        x_pos_ids[:, :, 0] = cap_feats.shape[1] + 1
+        x_pos_ids[:, :, 1] = torch.arange(H_tokens, dtype=torch.float32, device=device).view(-1, 1).repeat(1, W_tokens).flatten()
+        x_pos_ids[:, :, 2] = torch.arange(W_tokens, dtype=torch.float32, device=device).view(1, -1).repeat(H_tokens, 1).flatten()
 
-        position_ids = torch.zeros(bsz, max_seq_len, 3, dtype=torch.float32, device=device)
+        if self.pad_tokens_multiple is not None:
+            pad_extra = (-x.shape[1]) % self.pad_tokens_multiple
+            x = torch.cat((x, self.x_pad_token.to(device=x.device, dtype=x.dtype).unsqueeze(0).repeat(x.shape[0], pad_extra, 1)), dim=1)
+            x_pos_ids = torch.nn.functional.pad(x_pos_ids, (0, 0, 0, pad_extra))
 
-        for i in range(bsz):
-            cap_len = l_effective_cap_len[i]
-            img_len = l_effective_img_len[i]
-            H, W = img_sizes[i]
-            H_tokens, W_tokens = H // pH, W // pW
-            assert H_tokens * W_tokens == img_len
-
-            rope_options = transformer_options.get("rope_options", None)
-            h_scale = 1.0
-            w_scale = 1.0
-            h_start = 0
-            w_start = 0
-            if rope_options is not None:
-                h_scale = rope_options.get("scale_y", 1.0)
-                w_scale = rope_options.get("scale_x", 1.0)
-
-                h_start = rope_options.get("shift_y", 0.0)
-                w_start = rope_options.get("shift_x", 0.0)
-
-            position_ids[i, :cap_len, 0] = torch.arange(cap_len, dtype=torch.float32, device=device)
-            position_ids[i, cap_len:cap_len+img_len, 0] = cap_len
-            row_ids = (torch.arange(H_tokens, dtype=torch.float32, device=device) * h_scale + h_start).view(-1, 1).repeat(1, W_tokens).flatten()
-            col_ids = (torch.arange(W_tokens, dtype=torch.float32, device=device) * w_scale + w_start).view(1, -1).repeat(H_tokens, 1).flatten()
-            position_ids[i, cap_len:cap_len+img_len, 1] = row_ids
-            position_ids[i, cap_len:cap_len+img_len, 2] = col_ids
-
-        freqs_cis = self.rope_embedder(position_ids).movedim(1, 2).to(dtype)
-
-        # build freqs_cis for cap and image individually
-        cap_freqs_cis_shape = list(freqs_cis.shape)
-        # cap_freqs_cis_shape[1] = max_cap_len
-        cap_freqs_cis_shape[1] = cap_feats.shape[1]
-        cap_freqs_cis = torch.zeros(*cap_freqs_cis_shape, device=device, dtype=freqs_cis.dtype)
-
-        img_freqs_cis_shape = list(freqs_cis.shape)
-        img_freqs_cis_shape[1] = max_img_len
-        img_freqs_cis = torch.zeros(*img_freqs_cis_shape, device=device, dtype=freqs_cis.dtype)
-
-        for i in range(bsz):
-            cap_len = l_effective_cap_len[i]
-            img_len = l_effective_img_len[i]
-            cap_freqs_cis[i, :cap_len] = freqs_cis[i, :cap_len]
-            img_freqs_cis[i, :img_len] = freqs_cis[i, cap_len:cap_len+img_len]
+        freqs_cis = self.rope_embedder(torch.cat((cap_pos_ids, x_pos_ids), dim=1)).movedim(1, 2)
 
         # refine context
         for layer in self.context_refiner:
-            cap_feats = layer(cap_feats, cap_mask, cap_freqs_cis, transformer_options=transformer_options)
+            cap_feats = layer(cap_feats, cap_mask, freqs_cis[:, :cap_pos_ids.shape[1]], transformer_options=transformer_options)
 
-        # refine image
-        flat_x = []
-        for i in range(bsz):
-            img = x[i]
-            C, H, W = img.size()
-            img = img.view(C, H // pH, pH, W // pW, pW).permute(1, 3, 2, 4, 0).flatten(2).flatten(0, 1)
-            flat_x.append(img)
-        x = flat_x
-        padded_img_embed = torch.zeros(bsz, max_img_len, x[0].shape[-1], device=device, dtype=x[0].dtype)
-        padded_img_mask = torch.zeros(bsz, max_img_len, dtype=dtype, device=device)
-        for i in range(bsz):
-            padded_img_embed[i, :l_effective_img_len[i]] = x[i]
-            padded_img_mask[i, l_effective_img_len[i]:] = -torch.finfo(dtype).max
-
-        padded_img_embed = self.x_embedder(padded_img_embed)
-        padded_img_mask = padded_img_mask.unsqueeze(1)
+        padded_img_mask = None
         for layer in self.noise_refiner:
-            padded_img_embed = layer(padded_img_embed, padded_img_mask, img_freqs_cis, t, transformer_options=transformer_options)
-
-        if cap_mask is not None:
-            mask = torch.zeros(bsz, max_seq_len, dtype=dtype, device=device)
-            mask[:, :max_cap_len] = cap_mask[:, :max_cap_len]
-        else:
-            mask = None
-
-        padded_full_embed = torch.zeros(bsz, max_seq_len, self.dim, device=device, dtype=x[0].dtype)
-        for i in range(bsz):
-            cap_len = l_effective_cap_len[i]
-            img_len = l_effective_img_len[i]
-
-            padded_full_embed[i, :cap_len] = cap_feats[i, :cap_len]
-            padded_full_embed[i, cap_len:cap_len+img_len] = padded_img_embed[i, :img_len]
+            x = layer(x, padded_img_mask, freqs_cis[:, cap_pos_ids.shape[1]:], t, transformer_options=transformer_options)
 
+        padded_full_embed = torch.cat((cap_feats, x), dim=1)
+        mask = None
+        img_sizes = [(H, W)] * bsz
+        l_effective_cap_len = [cap_feats.shape[1]] * bsz
         return padded_full_embed, mask, img_sizes, l_effective_cap_len, freqs_cis
 
     def forward(self, x, timesteps, context, num_tokens, attention_mask=None, **kwargs):
@@ -627,7 +564,7 @@ class NextDiT(nn.Module):
         y: (N,) tensor of text tokens/features
         """
 
-        t = self.t_embedder(t, dtype=x.dtype)  # (N, D)
+        t = self.t_embedder(t * self.time_scale, dtype=x.dtype)  # (N, D)
         adaln_input = t
 
         cap_feats = self.cap_embedder(cap_feats)  # (N, L, D)  # todo check if able to batchify w.o. redundant compute

comfy/ldm/modules/diffusionmodules/mmdit.py

@@ -211,12 +211,14 @@ class TimestepEmbedder(nn.Module):
     Embeds scalar timesteps into vector representations.
     """
 
-    def __init__(self, hidden_size, frequency_embedding_size=256, dtype=None, device=None, operations=None):
+    def __init__(self, hidden_size, frequency_embedding_size=256, output_size=None, dtype=None, device=None, operations=None):
         super().__init__()
+        if output_size is None:
+            output_size = hidden_size
         self.mlp = nn.Sequential(
             operations.Linear(frequency_embedding_size, hidden_size, bias=True, dtype=dtype, device=device),
             nn.SiLU(),
-            operations.Linear(hidden_size, hidden_size, bias=True, dtype=dtype, device=device),
+            operations.Linear(hidden_size, output_size, bias=True, dtype=dtype, device=device),
         )
         self.frequency_embedding_size = frequency_embedding_size
 

comfy/model_base.py

@@ -1114,9 +1114,13 @@ class Lumina2(BaseModel):
             if torch.numel(attention_mask) != attention_mask.sum():
                 out['attention_mask'] = comfy.conds.CONDRegular(attention_mask)
             out['num_tokens'] = comfy.conds.CONDConstant(max(1, torch.sum(attention_mask).item()))
+
         cross_attn = kwargs.get("cross_attn", None)
         if cross_attn is not None:
             out['c_crossattn'] = comfy.conds.CONDRegular(cross_attn)
+            if 'num_tokens' not in out:
+                out['num_tokens'] = comfy.conds.CONDConstant(cross_attn.shape[1])
+
         return out
 
 class WAN21(BaseModel):

comfy/model_detection.py

@@ -416,14 +416,31 @@ def detect_unet_config(state_dict, key_prefix, metadata=None):
         dit_config["image_model"] = "lumina2"
         dit_config["patch_size"] = 2
         dit_config["in_channels"] = 16
-        dit_config["dim"] = 2304
-        dit_config["cap_feat_dim"] = state_dict['{}cap_embedder.1.weight'.format(key_prefix)].shape[1]
+        w = state_dict['{}cap_embedder.1.weight'.format(key_prefix)]
+        dit_config["dim"] = w.shape[0]
+        dit_config["cap_feat_dim"] = w.shape[1]
         dit_config["n_layers"] = count_blocks(state_dict_keys, '{}layers.'.format(key_prefix) + '{}.')
-        dit_config["n_heads"] = 24
-        dit_config["n_kv_heads"] = 8
         dit_config["qk_norm"] = True
-        dit_config["axes_dims"] = [32, 32, 32]
-        dit_config["axes_lens"] = [300, 512, 512]
+
+        if dit_config["dim"] == 2304: # Original Lumina 2
+            dit_config["n_heads"] = 24
+            dit_config["n_kv_heads"] = 8
+            dit_config["axes_dims"] = [32, 32, 32]
+            dit_config["axes_lens"] = [300, 512, 512]
+            dit_config["rope_theta"] = 10000.0
+            dit_config["ffn_dim_multiplier"] = 4.0
+        elif dit_config["dim"] == 3840:  # Z image
+            dit_config["n_heads"] = 30
+            dit_config["n_kv_heads"] = 30
+            dit_config["axes_dims"] = [32, 48, 48]
+            dit_config["axes_lens"] = [1536, 512, 512]
+            dit_config["rope_theta"] = 256.0
+            dit_config["ffn_dim_multiplier"] = (8.0 / 3.0)
+            dit_config["z_image_modulation"] = True
+            dit_config["time_scale"] = 1000.0
+            if '{}cap_pad_token'.format(key_prefix) in state_dict_keys:
+                dit_config["pad_tokens_multiple"] = 32
+
         return dit_config
 
     if '{}head.modulation'.format(key_prefix) in state_dict_keys:  # Wan 2.1

comfy/model_patcher.py

@@ -132,7 +132,7 @@ class LowVramPatch:
     def __call__(self, weight):
         intermediate_dtype = weight.dtype
         if self.convert_func is not None:
-            weight = self.convert_func(weight.to(dtype=torch.float32, copy=True), inplace=True)
+            weight = self.convert_func(weight, inplace=False)
 
         if intermediate_dtype not in [torch.float32, torch.float16, torch.bfloat16]: #intermediate_dtype has to be one that is supported in math ops
             intermediate_dtype = torch.float32

comfy/ops.py

@@ -117,6 +117,8 @@ def cast_bias_weight(s, input=None, dtype=None, device=None, bias_dtype=None, of
     if weight_has_function or weight.dtype != dtype:
         with wf_context:
             weight = weight.to(dtype=dtype)
+            if isinstance(weight, QuantizedTensor):
+                weight = weight.dequantize()
             for f in s.weight_function:
                 weight = f(weight)
 
@@ -502,7 +504,7 @@ def scaled_fp8_ops(fp8_matrix_mult=False, scale_input=False, override_dtype=None
                     weight *= self.scale_weight.to(device=weight.device, dtype=weight.dtype)
                     return weight
                 else:
-                    return weight * self.scale_weight.to(device=weight.device, dtype=weight.dtype)
+                    return weight.to(dtype=torch.float32) * self.scale_weight.to(device=weight.device, dtype=torch.float32)
 
             def set_weight(self, weight, inplace_update=False, seed=None, return_weight=False, **kwargs):
                 weight = comfy.float.stochastic_rounding(weight / self.scale_weight.to(device=weight.device, dtype=weight.dtype), self.weight.dtype, seed=seed)
@@ -643,6 +645,24 @@ def mixed_precision_ops(layer_quant_config={}, compute_dtype=torch.bfloat16, ful
                     not isinstance(input, QuantizedTensor)):
                     input = QuantizedTensor.from_float(input, self.layout_type, scale=self.input_scale, dtype=self.weight.dtype)
                 return self._forward(input, self.weight, self.bias)
+
+            def convert_weight(self, weight, inplace=False, **kwargs):
+                if isinstance(weight, QuantizedTensor):
+                    return weight.dequantize()
+                else:
+                    return weight
+
+            def set_weight(self, weight, inplace_update=False, seed=None, return_weight=False, **kwargs):
+                if getattr(self, 'layout_type', None) is not None:
+                    weight = QuantizedTensor.from_float(weight, self.layout_type, scale=None, dtype=self.weight.dtype, stochastic_rounding=seed, inplace_ops=True)
+                else:
+                    weight = weight.to(self.weight.dtype)
+                if return_weight:
+                    return weight
+
+                assert inplace_update is False  # TODO: eventually remove the inplace_update stuff
+                self.weight = torch.nn.Parameter(weight, requires_grad=False)
+
     return MixedPrecisionOps
 
 def pick_operations(weight_dtype, compute_dtype, load_device=None, disable_fast_fp8=False, fp8_optimizations=False, scaled_fp8=None, model_config=None):

comfy/quant_ops.py

@@ -1,6 +1,7 @@
 import torch
 import logging
 from typing import Tuple, Dict
+import comfy.float
 
 _LAYOUT_REGISTRY = {}
 _GENERIC_UTILS = {}
@@ -393,7 +394,7 @@ class TensorCoreFP8Layout(QuantizedLayout):
     - orig_dtype: Original dtype before quantization (for casting back)
     """
     @classmethod
-    def quantize(cls, tensor, scale=None, dtype=torch.float8_e4m3fn):
+    def quantize(cls, tensor, scale=None, dtype=torch.float8_e4m3fn, stochastic_rounding=0, inplace_ops=False):
         orig_dtype = tensor.dtype
 
         if scale is None:
@@ -403,17 +404,23 @@ class TensorCoreFP8Layout(QuantizedLayout):
             scale = torch.tensor(scale)
         scale = scale.to(device=tensor.device, dtype=torch.float32)
 
-        tensor_scaled = tensor * (1.0 / scale).to(tensor.dtype)
-        # TODO: uncomment this if it's actually needed because the clamp has a small performance penality'
-        lp_amax = torch.finfo(dtype).max
-        torch.clamp(tensor_scaled, min=-lp_amax, max=lp_amax, out=tensor_scaled)
-        qdata = tensor_scaled.to(dtype, memory_format=torch.contiguous_format)
+        if inplace_ops:
+            tensor *= (1.0 / scale).to(tensor.dtype)
+        else:
+            tensor = tensor * (1.0 / scale).to(tensor.dtype)
+
+        if stochastic_rounding > 0:
+            tensor = comfy.float.stochastic_rounding(tensor, dtype=dtype, seed=stochastic_rounding)
+        else:
+            lp_amax = torch.finfo(dtype).max
+            torch.clamp(tensor, min=-lp_amax, max=lp_amax, out=tensor)
+            tensor = tensor.to(dtype, memory_format=torch.contiguous_format)
 
         layout_params = {
             'scale': scale,
             'orig_dtype': orig_dtype
         }
-        return qdata, layout_params
+        return tensor, layout_params
 
     @staticmethod
     def dequantize(qdata, scale, orig_dtype, **kwargs):

comfy/sd.py

@@ -52,6 +52,7 @@ import comfy.text_encoders.ace
 import comfy.text_encoders.omnigen2
 import comfy.text_encoders.qwen_image
 import comfy.text_encoders.hunyuan_image
+import comfy.text_encoders.z_image
 
 import comfy.model_patcher
 import comfy.lora
@@ -953,6 +954,8 @@ class TEModel(Enum):
     GEMMA_3_4B = 13
     MISTRAL3_24B = 14
     MISTRAL3_24B_PRUNED_FLUX2 = 15
+    QWEN3_4B = 16
+
 
 def detect_te_model(sd):
     if "text_model.encoder.layers.30.mlp.fc1.weight" in sd:
@@ -985,6 +988,8 @@ def detect_te_model(sd):
         if weight.shape[0] == 512:
             return TEModel.QWEN25_7B
     if "model.layers.0.post_attention_layernorm.weight" in sd:
+        if 'model.layers.0.self_attn.q_norm.weight' in sd:
+            return TEModel.QWEN3_4B
         weight = sd['model.layers.0.post_attention_layernorm.weight']
         if weight.shape[0] == 5120:
             if "model.layers.39.post_attention_layernorm.weight" in sd:
@@ -1110,6 +1115,9 @@ def load_text_encoder_state_dicts(state_dicts=[], embedding_directory=None, clip
             clip_target.clip = comfy.text_encoders.flux.flux2_te(**llama_detect(clip_data), pruned=te_model == TEModel.MISTRAL3_24B_PRUNED_FLUX2)
             clip_target.tokenizer = comfy.text_encoders.flux.Flux2Tokenizer
             tokenizer_data["tekken_model"] = clip_data[0].get("tekken_model", None)
+        elif te_model == TEModel.QWEN3_4B:
+            clip_target.clip = comfy.text_encoders.z_image.te(**llama_detect(clip_data))
+            clip_target.tokenizer = comfy.text_encoders.z_image.ZImageTokenizer
         else:
             # clip_l
             if clip_type == CLIPType.SD3:

comfy/sd1_clip.py

@@ -90,7 +90,6 @@ class SDClipModel(torch.nn.Module, ClipTokenWeightEncoder):
                  special_tokens={"start": 49406, "end": 49407, "pad": 49407}, layer_norm_hidden_state=True, enable_attention_masks=False, zero_out_masked=False,
                  return_projected_pooled=True, return_attention_masks=False, model_options={}):  # clip-vit-base-patch32
         super().__init__()
-        assert layer in self.LAYERS
 
         if textmodel_json_config is None:
             textmodel_json_config = os.path.join(os.path.dirname(os.path.realpath(__file__)), "sd1_clip_config.json")
@@ -164,7 +163,7 @@ class SDClipModel(torch.nn.Module, ClipTokenWeightEncoder):
     def set_clip_options(self, options):
         layer_idx = options.get("layer", self.layer_idx)
         self.return_projected_pooled = options.get("projected_pooled", self.return_projected_pooled)
-        if self.layer == "all":
+        if isinstance(self.layer, list) or self.layer == "all":
             pass
         elif layer_idx is None or abs(layer_idx) > self.num_layers:
             self.layer = "last"
@@ -266,7 +265,9 @@ class SDClipModel(torch.nn.Module, ClipTokenWeightEncoder):
         if self.enable_attention_masks:
             attention_mask_model = attention_mask
 
-        if self.layer == "all":
+        if isinstance(self.layer, list):
+            intermediate_output = self.layer
+        elif self.layer == "all":
             intermediate_output = "all"
         else:
             intermediate_output = self.layer_idx

comfy/supported_models.py

@@ -21,6 +21,7 @@ import comfy.text_encoders.ace
 import comfy.text_encoders.omnigen2
 import comfy.text_encoders.qwen_image
 import comfy.text_encoders.hunyuan_image
+import comfy.text_encoders.z_image
 
 from . import supported_models_base
 from . import latent_formats
@@ -994,7 +995,7 @@ class Lumina2(supported_models_base.BASE):
         "shift": 6.0,
     }
 
-    memory_usage_factor = 1.2
+    memory_usage_factor = 1.4
 
     unet_extra_config = {}
     latent_format = latent_formats.Flux
@@ -1013,6 +1014,24 @@ class Lumina2(supported_models_base.BASE):
         hunyuan_detect = comfy.text_encoders.hunyuan_video.llama_detect(state_dict, "{}gemma2_2b.transformer.".format(pref))
         return supported_models_base.ClipTarget(comfy.text_encoders.lumina2.LuminaTokenizer, comfy.text_encoders.lumina2.te(**hunyuan_detect))
 
+class ZImage(Lumina2):
+    unet_config = {
+        "image_model": "lumina2",
+        "dim": 3840,
+    }
+
+    sampling_settings = {
+        "multiplier": 1.0,
+        "shift": 3.0,
+    }
+
+    memory_usage_factor = 1.7
+
+    def clip_target(self, state_dict={}):
+        pref = self.text_encoder_key_prefix[0]
+        hunyuan_detect = comfy.text_encoders.hunyuan_video.llama_detect(state_dict, "{}qwen3_4b.transformer.".format(pref))
+        return supported_models_base.ClipTarget(comfy.text_encoders.z_image.ZImageTokenizer, comfy.text_encoders.z_image.te(**hunyuan_detect))
+
 class WAN21_T2V(supported_models_base.BASE):
     unet_config = {
         "image_model": "wan2.1",
@@ -1453,7 +1472,7 @@ class HunyuanVideo15_SR_Distilled(HunyuanVideo):
         hunyuan_detect = comfy.text_encoders.hunyuan_video.llama_detect(state_dict, "{}qwen25_7b.transformer.".format(pref))
         return supported_models_base.ClipTarget(comfy.text_encoders.hunyuan_video.HunyuanVideo15Tokenizer, comfy.text_encoders.hunyuan_image.te(**hunyuan_detect))
 
-models = [LotusD, Stable_Zero123, SD15_instructpix2pix, SD15, SD20, SD21UnclipL, SD21UnclipH, SDXL_instructpix2pix, SDXLRefiner, SDXL, SSD1B, KOALA_700M, KOALA_1B, Segmind_Vega, SD_X4Upscaler, Stable_Cascade_C, Stable_Cascade_B, SV3D_u, SV3D_p, SD3, StableAudio, AuraFlow, PixArtAlpha, PixArtSigma, HunyuanDiT, HunyuanDiT1, FluxInpaint, Flux, FluxSchnell, GenmoMochi, LTXV, HunyuanVideo15_SR_Distilled, HunyuanVideo15, HunyuanImage21Refiner, HunyuanImage21, HunyuanVideoSkyreelsI2V, HunyuanVideoI2V, HunyuanVideo, CosmosT2V, CosmosI2V, CosmosT2IPredict2, CosmosI2VPredict2, Lumina2, WAN22_T2V, WAN21_T2V, WAN21_I2V, WAN21_FunControl2V, WAN21_Vace, WAN21_Camera, WAN22_Camera, WAN22_S2V, WAN21_HuMo, WAN22_Animate, Hunyuan3Dv2mini, Hunyuan3Dv2, Hunyuan3Dv2_1, HiDream, Chroma, ChromaRadiance, ACEStep, Omnigen2, QwenImage, Flux2]
+models = [LotusD, Stable_Zero123, SD15_instructpix2pix, SD15, SD20, SD21UnclipL, SD21UnclipH, SDXL_instructpix2pix, SDXLRefiner, SDXL, SSD1B, KOALA_700M, KOALA_1B, Segmind_Vega, SD_X4Upscaler, Stable_Cascade_C, Stable_Cascade_B, SV3D_u, SV3D_p, SD3, StableAudio, AuraFlow, PixArtAlpha, PixArtSigma, HunyuanDiT, HunyuanDiT1, FluxInpaint, Flux, FluxSchnell, GenmoMochi, LTXV, HunyuanVideo15_SR_Distilled, HunyuanVideo15, HunyuanImage21Refiner, HunyuanImage21, HunyuanVideoSkyreelsI2V, HunyuanVideoI2V, HunyuanVideo, CosmosT2V, CosmosI2V, CosmosT2IPredict2, CosmosI2VPredict2, ZImage, Lumina2, WAN22_T2V, WAN21_T2V, WAN21_I2V, WAN21_FunControl2V, WAN21_Vace, WAN21_Camera, WAN22_Camera, WAN22_S2V, WAN21_HuMo, WAN22_Animate, Hunyuan3Dv2mini, Hunyuan3Dv2, Hunyuan3Dv2_1, HiDream, Chroma, ChromaRadiance, ACEStep, Omnigen2, QwenImage, Flux2]
 
 
 models += [SVD_img2vid]

comfy/text_encoders/flux.py

@@ -87,6 +87,7 @@ def load_mistral_tokenizer(data):
     vocab = {}
 
     max_vocab = mistral_vocab["config"]["default_vocab_size"]
+    max_vocab -= len(mistral_vocab["special_tokens"])
 
     for w in mistral_vocab["vocab"]:
         r = w["rank"]
@@ -137,7 +138,7 @@ class Flux2Tokenizer(sd1_clip.SD1Tokenizer):
         return tokens
 
 class Mistral3_24BModel(sd1_clip.SDClipModel):
-    def __init__(self, device="cpu", layer="all", layer_idx=None, dtype=None, attention_mask=True, model_options={}):
+    def __init__(self, device="cpu", layer=[10, 20, 30], layer_idx=None, dtype=None, attention_mask=True, model_options={}):
         textmodel_json_config = {}
         num_layers = model_options.get("num_layers", None)
         if num_layers is not None:
@@ -153,7 +154,7 @@ class Flux2TEModel(sd1_clip.SD1ClipModel):
     def encode_token_weights(self, token_weight_pairs):
         out, pooled, extra = super().encode_token_weights(token_weight_pairs)
 
-        out = torch.stack((out[:, 10], out[:, 20], out[:, 30]), dim=1)
+        out = torch.stack((out[:, 0], out[:, 1], out[:, 2]), dim=1)
         out = out.movedim(1, 2)
         out = out.reshape(out.shape[0], out.shape[1], -1)
         return out, pooled, extra

comfy/text_encoders/llama.py

@@ -78,6 +78,28 @@ class Qwen25_3BConfig:
     rope_scale = None
     final_norm: bool = True
 
+@dataclass
+class Qwen3_4BConfig:
+    vocab_size: int = 151936
+    hidden_size: int = 2560
+    intermediate_size: int = 9728
+    num_hidden_layers: int = 36
+    num_attention_heads: int = 32
+    num_key_value_heads: int = 8
+    max_position_embeddings: int = 40960
+    rms_norm_eps: float = 1e-6
+    rope_theta: float = 1000000.0
+    transformer_type: str = "llama"
+    head_dim = 128
+    rms_norm_add = False
+    mlp_activation = "silu"
+    qkv_bias = False
+    rope_dims = None
+    q_norm = "gemma3"
+    k_norm = "gemma3"
+    rope_scale = None
+    final_norm: bool = True
+
 @dataclass
 class Qwen25_7BVLI_Config:
     vocab_size: int = 152064
@@ -434,8 +456,12 @@ class Llama2_(nn.Module):
 
         intermediate = None
         all_intermediate = None
+        only_layers = None
         if intermediate_output is not None:
-            if intermediate_output == "all":
+            if isinstance(intermediate_output, list):
+                all_intermediate = []
+                only_layers = set(intermediate_output)
+            elif intermediate_output == "all":
                 all_intermediate = []
                 intermediate_output = None
             elif intermediate_output < 0:
@@ -443,7 +469,8 @@ class Llama2_(nn.Module):
 
         for i, layer in enumerate(self.layers):
             if all_intermediate is not None:
-                all_intermediate.append(x.unsqueeze(1).clone())
+                if only_layers is None or (i in only_layers):
+                    all_intermediate.append(x.unsqueeze(1).clone())
             x = layer(
                 x=x,
                 attention_mask=mask,
@@ -457,7 +484,8 @@ class Llama2_(nn.Module):
             x = self.norm(x)
 
         if all_intermediate is not None:
-            all_intermediate.append(x.unsqueeze(1).clone())
+            if only_layers is None or ((i + 1) in only_layers):
+                all_intermediate.append(x.unsqueeze(1).clone())
 
         if all_intermediate is not None:
             intermediate = torch.cat(all_intermediate, dim=1)
@@ -505,6 +533,15 @@ class Qwen25_3B(BaseLlama, torch.nn.Module):
         self.model = Llama2_(config, device=device, dtype=dtype, ops=operations)
         self.dtype = dtype
 
+class Qwen3_4B(BaseLlama, torch.nn.Module):
+    def __init__(self, config_dict, dtype, device, operations):
+        super().__init__()
+        config = Qwen3_4BConfig(**config_dict)
+        self.num_layers = config.num_hidden_layers
+
+        self.model = Llama2_(config, device=device, dtype=dtype, ops=operations)
+        self.dtype = dtype
+
 class Qwen25_7BVLI(BaseLlama, torch.nn.Module):
     def __init__(self, config_dict, dtype, device, operations):
         super().__init__()

comfy/text_encoders/z_image.py

@@ -0,0 +1,48 @@
+from transformers import Qwen2Tokenizer
+import comfy.text_encoders.llama
+from comfy import sd1_clip
+import os
+
+class Qwen3Tokenizer(sd1_clip.SDTokenizer):
+    def __init__(self, embedding_directory=None, tokenizer_data={}):
+        tokenizer_path = os.path.join(os.path.dirname(os.path.realpath(__file__)), "qwen25_tokenizer")
+        super().__init__(tokenizer_path, pad_with_end=False, embedding_size=2560, embedding_key='qwen3_4b', tokenizer_class=Qwen2Tokenizer, has_start_token=False, has_end_token=False, pad_to_max_length=False, max_length=99999999, min_length=1, pad_token=151643, tokenizer_data=tokenizer_data)
+
+
+class ZImageTokenizer(sd1_clip.SD1Tokenizer):
+    def __init__(self, embedding_directory=None, tokenizer_data={}):
+        super().__init__(embedding_directory=embedding_directory, tokenizer_data=tokenizer_data, name="qwen3_4b", tokenizer=Qwen3Tokenizer)
+        self.llama_template = "<|im_start|>user\n{}<|im_end|>\n<|im_start|>assistant\n"
+
+    def tokenize_with_weights(self, text, return_word_ids=False, llama_template=None, **kwargs):
+        if llama_template is None:
+            llama_text = self.llama_template.format(text)
+        else:
+            llama_text = llama_template.format(text)
+
+        tokens = super().tokenize_with_weights(llama_text, return_word_ids=return_word_ids, disable_weights=True, **kwargs)
+        return tokens
+
+
+class Qwen3_4BModel(sd1_clip.SDClipModel):
+    def __init__(self, device="cpu", layer="hidden", layer_idx=-2, dtype=None, attention_mask=True, model_options={}):
+        super().__init__(device=device, layer=layer, layer_idx=layer_idx, textmodel_json_config={}, dtype=dtype, special_tokens={"pad": 151643}, layer_norm_hidden_state=False, model_class=comfy.text_encoders.llama.Qwen3_4B, enable_attention_masks=attention_mask, return_attention_masks=attention_mask, model_options=model_options)
+
+
+class ZImageTEModel(sd1_clip.SD1ClipModel):
+    def __init__(self, device="cpu", dtype=None, model_options={}):
+        super().__init__(device=device, dtype=dtype, name="qwen3_4b", clip_model=Qwen3_4BModel, model_options=model_options)
+
+
+def te(dtype_llama=None, llama_scaled_fp8=None, llama_quantization_metadata=None):
+    class ZImageTEModel_(ZImageTEModel):
+        def __init__(self, device="cpu", dtype=None, model_options={}):
+            if llama_scaled_fp8 is not None and "scaled_fp8" not in model_options:
+                model_options = model_options.copy()
+                model_options["scaled_fp8"] = llama_scaled_fp8
+            if dtype_llama is not None:
+                dtype = dtype_llama
+            if llama_quantization_metadata is not None:
+                model_options["quantization_metadata"] = llama_quantization_metadata
+            super().__init__(device=device, dtype=dtype, model_options=model_options)
+    return ZImageTEModel_

comfy/weight_adapter/lora.py

@@ -194,6 +194,7 @@ class LoRAAdapter(WeightAdapterBase):
             lora_diff = torch.mm(
                 mat1.flatten(start_dim=1), mat2.flatten(start_dim=1)
             ).reshape(weight.shape)
+            del mat1, mat2
             if dora_scale is not None:
                 weight = weight_decompose(
                     dora_scale,

comfyui_version.py

@@ -1,3 +1,3 @@
 # This file is automatically generated by the build process when version is
 # updated in pyproject.toml.
-__version__ = "0.3.72"
+__version__ = "0.3.74"

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "ComfyUI"
-version = "0.3.72"
+version = "0.3.74"
 readme = "README.md"
 license = { file = "LICENSE" }
 requires-python = ">=3.9"

requirements.txt

@@ -1,5 +1,5 @@
 comfyui-frontend-package==1.30.6
-comfyui-workflow-templates==0.7.9
+comfyui-workflow-templates==0.7.20
 comfyui-embedded-docs==0.3.1
 torch
 torchsde