Merge c0dc071f3b into 51bf508a0b

Signed-off-by: bigcat88 <bigcat88@icloud.com>

AGENTS.md

@@ -127,8 +127,6 @@
 - Do not add unnecessary `try`/`except` blocks. Use them for optional dependency,
   platform, or backend capability detection only when the program has a useful
   fallback. Prefer specific exception types when changing new code.
-- If a library version is pinned in `requirements.txt`, do not add code to
-  ComfyUI to handle older versions of that library.
 - Remove any workarounds for PyTorch versions that ComfyUI no longer officially
   supports. Deprecated workarounds include catching an exception and rerunning
   the same op with the input cast to float. If a workaround does not have a

README.md

@@ -229,7 +229,7 @@ Python 3.14 works but some custom nodes may have issues. The free threaded varia
 
 Python 3.13 is very well supported. If you have trouble with some custom node dependencies on 3.13 you can try 3.12
 
-torch 2.5 is minimally supported but using a newer version is extremely recommended. Some features and optimizations might only work on newer versions. We generally recommend using the latest major version of pytorch with the latest cuda version unless it is less than 2 weeks old. If your pytorch is more than 6 months old, please update it.
+torch 2.4 and above is supported but some features and optimizations might only work on newer versions. We generally recommend using the latest major version of pytorch with the latest cuda version unless it is less than 2 weeks old.
 
 ### Instructions:
 

comfy/ldm/ace/ace_step15.py

@@ -217,7 +217,10 @@ class AceStepAttention(nn.Module):
                 cos, sin = position_embeddings
                 query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin)
 
-        gqa_kwargs = {"enable_gqa": True} if self.num_heads != self.num_kv_heads else {}
+        n_rep = self.num_heads // self.num_kv_heads
+        if n_rep > 1:
+            key_states = key_states.repeat_interleave(n_rep, dim=1)
+            value_states = value_states.repeat_interleave(n_rep, dim=1)
 
         attn_bias = None
         if self.sliding_window is not None and not self.is_cross_attention:
@@ -241,7 +244,7 @@ class AceStepAttention(nn.Module):
             else:
                 attn_bias = window_bias
 
-        attn_output = optimized_attention(query_states, key_states, value_states, self.num_heads, attn_bias, skip_reshape=True, low_precision_attention=False, **gqa_kwargs)
+        attn_output = optimized_attention(query_states, key_states, value_states, self.num_heads, attn_bias, skip_reshape=True, low_precision_attention=False)
         attn_output = self.o_proj(attn_output)
 
         return attn_output

comfy/ldm/audio/dit.py

@@ -425,16 +425,19 @@ class Attention(nn.Module):
         if n == 1 and causal:
             causal = False
 
-        gqa_kwargs = {"enable_gqa": True} if h != kv_h else {}
+        if h != kv_h:
+            # Repeat interleave kv_heads to match q_heads
+            heads_per_kv_head = h // kv_h
+            k, v = map(lambda t: t.repeat_interleave(heads_per_kv_head, dim = 1), (k, v))
 
         if self.differential:
             q, q_diff = q.unbind(dim=1)
             k, k_diff = k.unbind(dim=1)
-            out      = optimized_attention(q,      k,      v, h, skip_reshape=True, low_precision_attention=False, transformer_options=transformer_options, **gqa_kwargs)
-            out_diff = optimized_attention(q_diff, k_diff, v, h, skip_reshape=True, low_precision_attention=False, transformer_options=transformer_options, **gqa_kwargs)
+            out      = optimized_attention(q,      k,      v, h, skip_reshape=True, low_precision_attention=False, transformer_options=transformer_options)
+            out_diff = optimized_attention(q_diff, k_diff, v, h, skip_reshape=True, low_precision_attention=False, transformer_options=transformer_options)
             out = out - out_diff
         else:
-            out = optimized_attention(q, k, v, h, skip_reshape=True, low_precision_attention=False, transformer_options=transformer_options, **gqa_kwargs)
+            out = optimized_attention(q, k, v, h, skip_reshape=True, low_precision_attention=False, transformer_options=transformer_options)
 
         out = self.to_out(out)
 

comfy/ldm/boogu/model.py

@@ -74,8 +74,11 @@ class BooguDoubleStreamProcessor(nn.Module):
         key = key.transpose(1, 2)
         value = value.transpose(1, 2)
 
-        gqa_kwargs = {"enable_gqa": True} if attn.kv_heads < attn.heads else {}
-        hidden_states = optimized_attention_masked(query, key, value, attn.heads, attention_mask, skip_reshape=True, transformer_options=transformer_options, **gqa_kwargs)
+        if attn.kv_heads < attn.heads:
+            key = key.repeat_interleave(attn.heads // attn.kv_heads, dim=1)
+            value = value.repeat_interleave(attn.heads // attn.kv_heads, dim=1)
+
+        hidden_states = optimized_attention_masked(query, key, value, attn.heads, attention_mask, skip_reshape=True, transformer_options=transformer_options)
 
         # Split back to instruction/image, apply per-stream output projections, recombine.
         instruct_hidden_states = self.instruct_out(hidden_states[:, :L_instruct])

comfy/ldm/modules/attention.py

@@ -1,6 +1,5 @@
 import math
 import sys
-import inspect
 
 import torch
 import torch.nn.functional as F
@@ -15,16 +14,16 @@ from .sub_quadratic_attention import efficient_dot_product_attention
 
 from comfy import model_management
 
+TORCH_HAS_GQA = model_management.torch_version_numeric >= (2, 5)
+
 if model_management.xformers_enabled():
     import xformers
     import xformers.ops
 
 SAGE_ATTENTION_IS_AVAILABLE = False
-SAGE_ATTENTION_SUPPORTS_MASK = False
 try:
     from sageattention import sageattn
     SAGE_ATTENTION_IS_AVAILABLE = True
-    SAGE_ATTENTION_SUPPORTS_MASK = "attn_mask" in inspect.signature(sageattn).parameters
 except ImportError as e:
     if model_management.sage_attention_enabled():
         if e.name == "sageattention":
@@ -90,44 +89,6 @@ def default(val, d):
         return val
     return d
 
-def _gqa_repeat_factor(query_heads, key_heads, value_heads):
-    if key_heads != value_heads:
-        raise ValueError(f"Key/value head count mismatch for GQA: {key_heads} != {value_heads}")
-    if query_heads == key_heads:
-        return 1
-    if query_heads % key_heads != 0:
-        raise ValueError(f"Query heads must be divisible by key/value heads for GQA: {query_heads} vs {key_heads}")
-    return query_heads // key_heads
-
-def _repeat_kv_for_gqa(k, v, query_heads, head_dim):
-    n_rep = _gqa_repeat_factor(query_heads, k.shape[head_dim], v.shape[head_dim])
-    if n_rep > 1:
-        k = k.repeat_interleave(n_rep, dim=head_dim)
-        v = v.repeat_interleave(n_rep, dim=head_dim)
-    return k, v
-
-def _heads_from_dim(tensor, dim_head, name):
-    inner_dim = tensor.shape[-1]
-    if inner_dim % dim_head != 0:
-        raise ValueError(f"{name} inner dimension {inner_dim} is not divisible by head dimension {dim_head}")
-    return inner_dim // dim_head
-
-def _reshape_qkv_to_heads(q, k, v, b, heads, dim_head, enable_gqa=False, expand_kv=True):
-    q = q.unsqueeze(3).reshape(b, -1, heads, dim_head)
-    if enable_gqa:
-        key_heads = _heads_from_dim(k, dim_head, "Key")
-        value_heads = _heads_from_dim(v, dim_head, "Value")
-    else:
-        key_heads = heads
-        value_heads = heads
-    k = k.unsqueeze(3).reshape(b, -1, key_heads, dim_head)
-    v = v.unsqueeze(3).reshape(b, -1, value_heads, dim_head)
-    if enable_gqa:
-        _gqa_repeat_factor(heads, key_heads, value_heads)
-        if expand_kv:
-            k, v = _repeat_kv_for_gqa(k, v, heads, -2)
-    return q, k, v
-
 
 # feedforward
 class GEGLU(nn.Module):
@@ -191,19 +152,28 @@ def attention_basic(q, k, v, heads, mask=None, attn_precision=None, skip_reshape
         b, _, dim_head = q.shape
         dim_head //= heads
 
+    if kwargs.get("enable_gqa", False) and q.shape[-3] != k.shape[-3]:
+        n_rep = q.shape[-3] // k.shape[-3]
+        k = k.repeat_interleave(n_rep, dim=-3)
+        v = v.repeat_interleave(n_rep, dim=-3)
+
     scale = kwargs.get("scale", dim_head ** -0.5)
 
     h = heads
     if skip_reshape:
-        if kwargs.get("enable_gqa", False):
-            k, v = _repeat_kv_for_gqa(k, v, q.shape[-3], -3)
-        q, k, v = map(
+         q, k, v = map(
             lambda t: t.reshape(b * heads, -1, dim_head),
             (q, k, v),
         )
     else:
-        q, k, v = _reshape_qkv_to_heads(q, k, v, b, heads, dim_head, kwargs.get("enable_gqa", False))
-        q, k, v = map(lambda t: t.permute(0, 2, 1, 3).reshape(b * heads, -1, dim_head).contiguous(), (q, k, v))
+        q, k, v = map(
+            lambda t: t.unsqueeze(3)
+            .reshape(b, -1, heads, dim_head)
+            .permute(0, 2, 1, 3)
+            .reshape(b * heads, -1, dim_head)
+            .contiguous(),
+            (q, k, v),
+        )
 
     # force cast to fp32 to avoid overflowing
     if attn_precision == torch.float32:
@@ -261,16 +231,13 @@ def attention_sub_quad(query, key, value, heads, mask=None, attn_precision=None,
         query = query * (kwargs["scale"] * dim_head ** 0.5)
 
     if skip_reshape:
-        if kwargs.get("enable_gqa", False):
-            key, value = _repeat_kv_for_gqa(key, value, query.shape[-3], -3)
         query = query.reshape(b * heads, -1, dim_head)
         value = value.reshape(b * heads, -1, dim_head)
         key = key.reshape(b * heads, -1, dim_head).movedim(1, 2)
     else:
-        query, key, value = _reshape_qkv_to_heads(query, key, value, b, heads, dim_head, kwargs.get("enable_gqa", False))
-        query = query.permute(0, 2, 1, 3).reshape(b * heads, -1, dim_head)
-        value = value.permute(0, 2, 1, 3).reshape(b * heads, -1, dim_head)
-        key = key.permute(0, 2, 3, 1).reshape(b * heads, dim_head, -1)
+        query = query.unsqueeze(3).reshape(b, -1, heads, dim_head).permute(0, 2, 1, 3).reshape(b * heads, -1, dim_head)
+        value = value.unsqueeze(3).reshape(b, -1, heads, dim_head).permute(0, 2, 1, 3).reshape(b * heads, -1, dim_head)
+        key = key.unsqueeze(3).reshape(b, -1, heads, dim_head).permute(0, 2, 3, 1).reshape(b * heads, dim_head, -1)
 
 
     dtype = query.dtype
@@ -337,15 +304,19 @@ def attention_split(q, k, v, heads, mask=None, attn_precision=None, skip_reshape
     scale = kwargs.get("scale", dim_head ** -0.5)
 
     if skip_reshape:
-        if kwargs.get("enable_gqa", False):
-            k, v = _repeat_kv_for_gqa(k, v, q.shape[-3], -3)
-        q, k, v = map(
+         q, k, v = map(
             lambda t: t.reshape(b * heads, -1, dim_head),
             (q, k, v),
         )
     else:
-        q, k, v = _reshape_qkv_to_heads(q, k, v, b, heads, dim_head, kwargs.get("enable_gqa", False))
-        q, k, v = map(lambda t: t.permute(0, 2, 1, 3).reshape(b * heads, -1, dim_head).contiguous(), (q, k, v))
+        q, k, v = map(
+            lambda t: t.unsqueeze(3)
+            .reshape(b, -1, heads, dim_head)
+            .permute(0, 2, 1, 3)
+            .reshape(b * heads, -1, dim_head)
+            .contiguous(),
+            (q, k, v),
+        )
 
     r1 = torch.zeros(q.shape[0], q.shape[1], v.shape[2], device=q.device, dtype=q.dtype)
 
@@ -467,7 +438,7 @@ def attention_xformers(q, k, v, heads, mask=None, attn_precision=None, skip_resh
             disabled_xformers = True
 
     if disabled_xformers:
-        return attention_pytorch(q, k, v, heads, mask, skip_reshape=skip_reshape, skip_output_reshape=skip_output_reshape, **kwargs)
+        return attention_pytorch(q, k, v, heads, mask, skip_reshape=skip_reshape, **kwargs)
 
     if skip_reshape:
         # b h k d -> b k h d
@@ -475,12 +446,13 @@ def attention_xformers(q, k, v, heads, mask=None, attn_precision=None, skip_resh
             lambda t: t.permute(0, 2, 1, 3),
             (q, k, v),
         )
-        if kwargs.get("enable_gqa", False):
-            k, v = _repeat_kv_for_gqa(k, v, q.shape[-2], -2)
     # actually do the reshaping
     else:
         dim_head //= heads
-        q, k, v = _reshape_qkv_to_heads(q, k, v, b, heads, dim_head, kwargs.get("enable_gqa", False))
+        q, k, v = map(
+            lambda t: t.reshape(b, -1, heads, dim_head),
+            (q, k, v),
+        )
 
     if mask is not None:
         # add a singleton batch dimension
@@ -502,7 +474,7 @@ def attention_xformers(q, k, v, heads, mask=None, attn_precision=None, skip_resh
         mask = mask_out[..., :mask.shape[-1]]
         mask = mask.expand(b, heads, -1, -1)
 
-    out = xformers.ops.memory_efficient_attention(q, k, v, attn_bias=mask, scale=kwargs.get("scale", None))
+    out = xformers.ops.memory_efficient_attention(q, k, v, attn_bias=mask)
 
     if skip_output_reshape:
         out = out.permute(0, 2, 1, 3)
@@ -526,8 +498,10 @@ def attention_pytorch(q, k, v, heads, mask=None, attn_precision=None, skip_resha
     else:
         b, _, dim_head = q.shape
         dim_head //= heads
-        q, k, v = _reshape_qkv_to_heads(q, k, v, b, heads, dim_head, kwargs.get("enable_gqa", False), expand_kv=False)
-        q, k, v = map(lambda t: t.transpose(1, 2), (q, k, v))
+        q, k, v = map(
+            lambda t: t.view(b, -1, heads, dim_head).transpose(1, 2),
+            (q, k, v),
+        )
 
     if mask is not None:
         # add a batch dimension if there isn't already one
@@ -537,7 +511,9 @@ def attention_pytorch(q, k, v, heads, mask=None, attn_precision=None, skip_resha
         if mask.ndim == 3:
             mask = mask.unsqueeze(1)
 
-    sdpa_keys = ("scale", "enable_gqa")
+    # Pass through extra SDPA kwargs (scale, enable_gqa) if provided
+    # enable_gqa requires PyTorch 2.5+; older versions use manual KV expansion above
+    sdpa_keys = ("scale", "enable_gqa") if TORCH_HAS_GQA else ("scale",)
     sdpa_extra = {k: v for k, v in kwargs.items() if k in sdpa_keys}
 
     if SDP_BATCH_LIMIT >= b:
@@ -565,19 +541,20 @@ def attention_pytorch(q, k, v, heads, mask=None, attn_precision=None, skip_resha
 
 @wrap_attn
 def attention_sage(q, k, v, heads, mask=None, attn_precision=None, skip_reshape=False, skip_output_reshape=False, **kwargs):
-    if kwargs.get("low_precision_attention", True) is False or (mask is not None and not SAGE_ATTENTION_SUPPORTS_MASK):
+    if kwargs.get("low_precision_attention", True) is False:
         return attention_pytorch(q, k, v, heads, mask=mask, skip_reshape=skip_reshape, skip_output_reshape=skip_output_reshape, **kwargs)
 
     exception_fallback = False
     if skip_reshape:
         b, _, _, dim_head = q.shape
         tensor_layout = "HND"
-        if kwargs.get("enable_gqa", False):
-            k, v = _repeat_kv_for_gqa(k, v, q.shape[-3], -3)
     else:
         b, _, dim_head = q.shape
         dim_head //= heads
-        q, k, v = _reshape_qkv_to_heads(q, k, v, b, heads, dim_head, kwargs.get("enable_gqa", False))
+        q, k, v = map(
+            lambda t: t.view(b, -1, heads, dim_head),
+            (q, k, v),
+        )
         tensor_layout = "NHD"
 
     if mask is not None:
@@ -588,12 +565,8 @@ def attention_sage(q, k, v, heads, mask=None, attn_precision=None, skip_reshape=
         if mask.ndim == 3:
             mask = mask.unsqueeze(1)
 
-    sage_kwargs = {"is_causal": False, "tensor_layout": tensor_layout, "sm_scale": kwargs.get("scale", None), "smooth_k": False}
-    if mask is not None:
-        sage_kwargs["attn_mask"] = mask
-
     try:
-        out = sageattn(q, k, v, **sage_kwargs)
+        out = sageattn(q, k, v, attn_mask=mask, is_causal=False, tensor_layout=tensor_layout)
     except Exception as e:
         logging.error("Error running sage attention: {}, using pytorch attention instead.".format(e))
         exception_fallback = True
@@ -643,6 +616,7 @@ def attention3_sage(q, k, v, heads, mask=None, attn_precision=None, skip_reshape
                 skip_output_reshape=skip_output_reshape,
                 **kwargs
             )
+        q_s, k_s, v_s = q, k, v
         N = q.shape[2]
         dim_head = D
     else:
@@ -668,15 +642,11 @@ def attention3_sage(q, k, v, heads, mask=None, attn_precision=None, skip_reshape
                 **kwargs
             )
 
-    if skip_reshape:
-        q_s = q
-        if kwargs.get("enable_gqa", False):
-            k_s, v_s = _repeat_kv_for_gqa(k, v, H, -3)
-        else:
-            k_s, v_s = k, v
-    else:
-        q_s, k_s, v_s = _reshape_qkv_to_heads(q, k, v, B, heads, dim_head, kwargs.get("enable_gqa", False))
-        q_s, k_s, v_s = map(lambda t: t.permute(0, 2, 1, 3).contiguous(), (q_s, k_s, v_s))
+    if not skip_reshape:
+        q_s, k_s, v_s = map(
+            lambda t: t.view(B, -1, heads, dim_head).permute(0, 2, 1, 3).contiguous(),
+            (q, k, v),
+        )
         B, H, L, D = q_s.shape
 
     try:
@@ -692,7 +662,7 @@ def attention3_sage(q, k, v, heads, mask=None, attn_precision=None, skip_reshape
                 q, k, v, heads,
                 mask=mask,
                 attn_precision=attn_precision,
-                skip_reshape=skip_reshape,
+                skip_reshape=False,
                 skip_output_reshape=skip_output_reshape,
                 **kwargs
             )
@@ -711,20 +681,19 @@ def attention3_sage(q, k, v, heads, mask=None, attn_precision=None, skip_reshape
 try:
     @torch.library.custom_op("flash_attention::flash_attn", mutates_args=())
     def flash_attn_wrapper(q: torch.Tensor, k: torch.Tensor, v: torch.Tensor,
-                    dropout_p: float = 0.0, causal: bool = False, softmax_scale: float = -1.0) -> torch.Tensor:
-        softmax_scale_arg = None if softmax_scale == -1.0 else softmax_scale
-        return flash_attn_func(q, k, v, dropout_p=dropout_p, causal=causal, softmax_scale=softmax_scale_arg)
+                    dropout_p: float = 0.0, causal: bool = False) -> torch.Tensor:
+        return flash_attn_func(q, k, v, dropout_p=dropout_p, causal=causal)
 
 
     @flash_attn_wrapper.register_fake
-    def flash_attn_fake(q, k, v, dropout_p=0.0, causal=False, softmax_scale=-1.0):
+    def flash_attn_fake(q, k, v, dropout_p=0.0, causal=False):
         # Output shape is the same as q
         return q.new_empty(q.shape)
 except AttributeError as error:
     FLASH_ATTN_ERROR = error
 
     def flash_attn_wrapper(q: torch.Tensor, k: torch.Tensor, v: torch.Tensor,
-                    dropout_p: float = 0.0, causal: bool = False, softmax_scale: float = -1.0) -> torch.Tensor:
+                    dropout_p: float = 0.0, causal: bool = False) -> torch.Tensor:
         assert False, f"Could not define flash_attn_wrapper: {FLASH_ATTN_ERROR}"
 
 @wrap_attn
@@ -734,8 +703,10 @@ def attention_flash(q, k, v, heads, mask=None, attn_precision=None, skip_reshape
     else:
         b, _, dim_head = q.shape
         dim_head //= heads
-        q, k, v = _reshape_qkv_to_heads(q, k, v, b, heads, dim_head, kwargs.get("enable_gqa", False), expand_kv=False)
-        q, k, v = map(lambda t: t.transpose(1, 2), (q, k, v))
+        q, k, v = map(
+            lambda t: t.view(b, -1, heads, dim_head).transpose(1, 2),
+            (q, k, v),
+        )
 
     if mask is not None:
         # add a batch dimension if there isn't already one
@@ -754,16 +725,10 @@ def attention_flash(q, k, v, heads, mask=None, attn_precision=None, skip_reshape
             v.transpose(1, 2),
             dropout_p=0.0,
             causal=False,
-            softmax_scale=kwargs.get("scale", -1.0),
         ).transpose(1, 2)
     except Exception as e:
         logging.warning(f"Flash Attention failed, using default SDPA: {e}")
-        sdpa_extra = {}
-        if kwargs.get("enable_gqa", False):
-            sdpa_extra["enable_gqa"] = True
-        if "scale" in kwargs:
-            sdpa_extra["scale"] = kwargs["scale"]
-        out = torch.nn.functional.scaled_dot_product_attention(q, k, v, attn_mask=mask, dropout_p=0.0, is_causal=False, **sdpa_extra)
+        out = torch.nn.functional.scaled_dot_product_attention(q, k, v, attn_mask=mask, dropout_p=0.0, is_causal=False)
     if not skip_output_reshape:
         out = (
             out.transpose(1, 2).reshape(b, -1, heads * dim_head)
@@ -1244,3 +1209,5 @@ class SpatialVideoTransformer(SpatialTransformer):
             x = self.proj_out(x)
         out = x + x_in
         return out
+
+

comfy/ldm/omnigen/omnigen2.py

@@ -141,8 +141,11 @@ class Attention(nn.Module):
         key = key.transpose(1, 2)
         value = value.transpose(1, 2)
 
-        gqa_kwargs = {"enable_gqa": True} if self.kv_heads < self.heads else {}
-        hidden_states = optimized_attention_masked(query, key, value, self.heads, attention_mask, skip_reshape=True, transformer_options=transformer_options, **gqa_kwargs)
+        if self.kv_heads < self.heads:
+            key = key.repeat_interleave(self.heads // self.kv_heads, dim=1)
+            value = value.repeat_interleave(self.heads // self.kv_heads, dim=1)
+
+        hidden_states = optimized_attention_masked(query, key, value, self.heads, attention_mask, skip_reshape=True, transformer_options=transformer_options)
         hidden_states = self.to_out[0](hidden_states)
         return hidden_states
 

comfy/ops.py

@@ -174,8 +174,6 @@ def cast_modules_with_vbar(comfy_modules, dtype, device, bias_dtype, non_blockin
                     elif xfer_dest2 is not None:
                         xfer_source.prepare(xfer_dest2, stream, copy=True, commit=False)
                         return
-                    else:
-                        return
                 comfy.model_management.cast_to_gathered(xfer_source, xfer_dest, non_blocking=non_blocking, stream=stream, r2=xfer_dest2)
 
         def handle_pin(m, pin, source, dest, subset="weights", size=None):

comfy/text_encoders/gpt_oss.py

@@ -12,7 +12,7 @@ import torch.nn.functional as F
 
 import comfy.ops
 from comfy import sd1_clip
-from comfy.ldm.modules.attention import optimized_attention_for_device
+from comfy.ldm.modules.attention import TORCH_HAS_GQA, optimized_attention_for_device
 from comfy.text_encoders.llama import RMSNorm, apply_rope
 
 
@@ -110,6 +110,10 @@ def _attention_with_sinks(q: torch.Tensor, k: torch.Tensor, v: torch.Tensor, sin
     putting the sink logit in the mask at that column.
     """
 
+    if num_kv_groups > 1 and not TORCH_HAS_GQA:
+        k = k.repeat_interleave(num_kv_groups, dim=1)
+        v = v.repeat_interleave(num_kv_groups, dim=1)
+
     B, _, S_q, D = q.shape
     H_kv = k.shape[1]
     S_kv = k.shape[-2]

comfy/text_encoders/llama.py

@@ -550,8 +550,10 @@ class Attention(nn.Module):
                 xv = xv[:, :, -sliding_window:]
                 attention_mask = attention_mask[..., -sliding_window:] if attention_mask is not None else None
 
-        gqa_kwargs = {"enable_gqa": True} if self.num_heads != self.num_kv_heads else {}
-        output = optimized_attention(xq, xk, xv, self.num_heads, mask=attention_mask, skip_reshape=True, **gqa_kwargs)
+        xk = xk.repeat_interleave(self.num_heads // self.num_kv_heads, dim=1)
+        xv = xv.repeat_interleave(self.num_heads // self.num_kv_heads, dim=1)
+
+        output = optimized_attention(xq, xk, xv, self.num_heads, mask=attention_mask, skip_reshape=True)
         return self.o_proj(output), present_key_value
 
 class MLP(nn.Module):

comfy/text_encoders/qwen35.py

@@ -366,8 +366,12 @@ class GatedAttention(nn.Module):
                     xv = torch.cat((past_value[:, :, :index], xv), dim=2)
                 present_key_value = (xk, xv, index + num_tokens)
 
-        gqa_kwargs = {"enable_gqa": True} if self.num_heads != self.num_kv_heads else {}
-        output = optimized_attention(xq, xk, xv, self.num_heads, mask=attention_mask, skip_reshape=True, **gqa_kwargs)
+        # Expand KV heads for GQA
+        if self.num_heads != self.num_kv_heads:
+            xk = xk.repeat_interleave(self.num_heads // self.num_kv_heads, dim=1)
+            xv = xv.repeat_interleave(self.num_heads // self.num_kv_heads, dim=1)
+
+        output = optimized_attention(xq, xk, xv, self.num_heads, mask=attention_mask, skip_reshape=True)
         output = output * gate.sigmoid()
 
         return self.o_proj(output), present_key_value

comfy_api/latest/_input_impl/video_types.py

@@ -556,9 +556,6 @@ class VideoFromFile(VideoInput):
         audio_done = audio_stream is None
         video_pts_offset = None
         last_video_pts = None
-        last_video_end = None
-        # rebased pts -> true display duration: the mp4 muxer pads the last sample with 1/rate otherwise
-        video_frame_durations = {}
         source_size = None
         rotation_k = 0
         rotation_filter = None
@@ -577,12 +574,13 @@ class VideoFromFile(VideoInput):
             return frame
 
         def drain_audio(final=False):
-            # Audio may cover the pts span of the video written so far, capped by the requested duration
+            # Audio may cover the pts span of the video written so far, capped by the requested duration;
+            # frames/average_rate would misjudge VFR spans.
             nonlocal samples_written, audio_done
-            if last_video_end is None:
+            if last_video_pts is None:
                 cap = 0
             else:
-                cap = math.ceil(last_video_end * video_stream.time_base * sample_rate)
+                cap = math.ceil((last_video_pts + pts_step) * video_stream.time_base * sample_rate)
             if duration_cap is not None:
                 cap = min(cap, duration_cap)
             while pending_audio and not audio_done:
@@ -624,15 +622,7 @@ class VideoFromFile(VideoInput):
                             continue
                         if end_pts is not None and frame.pts is not None and frame.pts >= end_pts:
                             video_done = True
-                            if last_video_pts is not None:
-                                # the source continues past the window: hold the last kept frame to the window end
-                                last_video_end = max(last_video_end, end_pts - video_pts_offset)
                             break
-                        # the source's true display duration of this frame; average_rate is not a
-                        # frame duration (sparse/VFR sources), so it is only the fallback
-                        frame_duration = frame.duration if frame.duration else pts_step
-                        if end_pts is not None and frame.pts is not None:
-                            frame_duration = min(frame_duration, end_pts - frame.pts)
                         if output is None:
                             rotation_k = int(round(frame.rotation // 90)) % 4 if frame.rotation else 0
                             if rotation_k % 2:
@@ -646,9 +636,6 @@ class VideoFromFile(VideoInput):
                             # Add metadata before writing any streams
                             write_output_metadata(container, output, metadata)
                             out_video = output.add_stream("h264", rate=rate)
-                            # no B-frames: reordering makes mp4 sample durations follow decode order,
-                            # so irregular-VFR spans and trim windows land wrong
-                            out_video.codec_context.max_b_frames = 0
                             out_video.width = out_width
                             out_video.height = out_height
                             out_video.pix_fmt = pix_fmt
@@ -694,14 +681,10 @@ class VideoFromFile(VideoInput):
                             # muxer rejects; nudge them forward by one nominal frame interval
                             frame.pts = 0 if last_video_pts is None else last_video_pts + pts_step
                         last_video_pts = frame.pts
-                        last_video_end = frame.pts + frame_duration
-                        video_frame_durations[frame.pts] = frame_duration
                         # the decoded pict_type would force x264's frame types (intra-only
                         # sources like MJPEG/ProRes would come out all-keyframe)
                         frame.pict_type = 0
-                        for out_packet in out_video.encode(frame):
-                            out_packet.duration = video_frame_durations.pop(out_packet.pts, 0)
-                            output.mux(out_packet)
+                        output.mux(out_video.encode(frame))
                         drain_audio()
 
                 elif packet.stream == audio_stream and not audio_done:
@@ -735,13 +718,7 @@ class VideoFromFile(VideoInput):
                 raise ValueError(f"No decodable video frames found in file '{self.__file}'")
             if out_audio is not None and not audio_done:
                 drain_audio(final=True)
-            window_fill = last_video_end - last_video_pts if video_done and last_video_pts is not None else 0
-            for out_packet in out_video.encode(None):
-                duration = video_frame_durations.pop(out_packet.pts, 0)
-                if out_packet.pts == last_video_pts:
-                    duration = max(duration, window_fill)
-                out_packet.duration = duration
-                output.mux(out_packet)
+            output.mux(out_video.encode(None))
             if out_audio is not None:
                 output.mux(out_audio.encode(None))
         except BaseException:

tests-unit/comfy_api_test/video_types_test.py

@@ -516,85 +516,6 @@ def test_save_to_transcode_trimmed_fragmented_mp4_keeps_audio():
     assert result["audio_seconds"] == pytest.approx(1.0, abs=0.05)
 
 
-def test_save_to_transcode_sparse_video_keeps_true_duration():
-    """average_rate is not a frame duration: a 3-frame video spanning 60 s averages
-    0.05 fps, and padding the last frame with 1/average_rate used to extend the
-    output — and the audio kept with it — about 20 s past the source span."""
-    sample_rate = 48000
-    buffer = io.BytesIO()
-    with av.open(buffer, mode="w", format="mp4") as container:
-        video_stream = container.add_stream("mpeg4", rate=30)
-        video_stream.width = video_stream.height = 64
-        video_stream.pix_fmt = "yuv420p"
-        audio_stream = container.add_stream("aac", rate=sample_rate, layout="mono")
-        for i, second in enumerate((0, 30, 60)):
-            frame = av.VideoFrame.from_ndarray(
-                torch.full((64, 64, 3), i * 80, dtype=torch.uint8).numpy(), format="rgb24"
-            ).reformat(format="yuv420p")
-            frame.pts = second * 30
-            frame.time_base = Fraction(1, 30)
-            container.mux(video_stream.encode(frame))
-        for offset in range(0, 90 * sample_rate, 1024):
-            n = min(1024, 90 * sample_rate - offset)
-            audio_frame = av.AudioFrame.from_ndarray(
-                torch.zeros(1, n, dtype=torch.float32).numpy(), format="fltp", layout="mono"
-            )
-            audio_frame.sample_rate = sample_rate
-            audio_frame.pts = offset
-            container.mux(audio_stream.encode(audio_frame))
-        for stream in (video_stream, audio_stream):
-            container.mux(stream.encode(None))
-
-    result = transcode_and_probe(VideoFromFile(buffer))
-    assert result["frames"] == 3
-    # the last frame keeps its true stts duration (1/30 s), not 1/average_rate (~20 s)
-    assert result["video_seconds"] == pytest.approx(60.03, abs=0.05)
-    assert result["audio_seconds"] == pytest.approx(60.03, abs=0.1)
-
-    trimmed = transcode_and_probe(VideoFromFile(buffer, duration=45))
-    assert trimmed["frames"] == 2
-    # a kept frame whose source duration crosses the window end is clamped to it
-    assert trimmed["video_seconds"] == pytest.approx(45.0, abs=0.05)
-    assert trimmed["audio_seconds"] == pytest.approx(45.0, abs=0.1)
-
-
-def test_save_to_transcode_irregular_vfr_keeps_span():
-    """B-frames reorder packets, and mp4 sample durations follow decode order: the dts
-    timeline ends before the pts timeline, so an irregular-VFR source's tail holds fell
-    out of the container (this 20.23 s span used to come out as 15.27 s, and the 10 s
-    trim as 6.03 s). The transcode encodes without B-frames so every sample keeps its
-    true display duration."""
-    durations = [1, 1, 60, 1, 1, 120, 1, 180, 1, 1, 150, 90]  # 1/30 s ticks, span 20.2333 s
-    generator = torch.Generator().manual_seed(7)
-    buffer = io.BytesIO()
-    with av.open(buffer, mode="w", format="mp4") as container:
-        video_stream = container.add_stream("mpeg4", rate=30)
-        video_stream.width = video_stream.height = 64
-        video_stream.pix_fmt = "yuv420p"
-        pts = 0
-        for duration in durations:
-            # textured frames, so an encoder with default settings has B-frames to gain from
-            frame = av.VideoFrame.from_ndarray(
-                torch.randint(0, 255, (64, 64, 3), generator=generator, dtype=torch.uint8).numpy(),
-                format="rgb24",
-            ).reformat(format="yuv420p")
-            frame.pts = pts
-            frame.time_base = Fraction(1, 30)
-            pts += duration
-            for packet in video_stream.encode(frame):
-                packet.duration = duration  # exact stts in the source
-                container.mux(packet)
-        container.mux(video_stream.encode(None))
-
-    result = transcode_and_probe(VideoFromFile(buffer))
-    assert result["frames"] == len(durations)
-    assert result["video_seconds"] == pytest.approx(sum(durations) / 30, abs=0.05)
-
-    trimmed = transcode_and_probe(VideoFromFile(buffer, duration=10))
-    assert trimmed["frames"] == 8  # frames at 12.167 s+ fall outside the window
-    assert trimmed["video_seconds"] == pytest.approx(10.0, abs=0.05)
-
-
 def test_save_to_transcode_bakes_rotation():
     """A 90-degree display-matrix rotation swaps the output dimensions (portrait video)"""
     file_path = create_transcode_source(width=64, height=32, rotation=True)