dynamic_vram: Fix windows Aimdo crash + Fix LLM performance (Comfy-Org#12408)

* model_management: lazy-cache aimdo_tensor

These tensors cosntructed from aimdo-allocations are CPU expensive to
make on the pytorch side. Add a cache version that will be valid with
signature match to fast path past whatever torch is doing.

* dynamic_vram: Minimize fast path CPU work

Move as much as possible inside the not resident if block and cache
the formed weight and bias rather than the flat intermediates. In
extreme layer weight rates this adds up.

Author: rattus128

comfy/model_management.py

@@ -1213,8 +1213,12 @@ def cast_to(weight, dtype=None, device=None, non_blocking=False, copy=False, str
 
         signature = comfy_aimdo.model_vbar.vbar_fault(weight._v)
         if signature is not None:
-            v_tensor = comfy.memory_management.interpret_gathered_like(cast_geometry, weight._v_tensor)[0]
-            if not comfy_aimdo.model_vbar.vbar_signature_compare(signature, weight._v_signature):
+            if comfy_aimdo.model_vbar.vbar_signature_compare(signature, weight._v_signature):
+                v_tensor = weight._v_tensor
+            else:
+                raw_tensor = comfy_aimdo.torch.aimdo_to_tensor(weight._v, device)
+                v_tensor = comfy.memory_management.interpret_gathered_like(cast_geometry, raw_tensor)[0]
+                weight._v_tensor = v_tensor
                 weight._v_signature = signature
                 #Send it over
                 v_tensor.copy_(weight, non_blocking=non_blocking)

comfy/model_patcher.py

@@ -1542,7 +1542,6 @@ def setup_param(self, m, n, param_key):
 
                     if vbar is not None and not hasattr(m, "_v"):
                         m._v = vbar.alloc(v_weight_size)
-                        m._v_tensor = comfy_aimdo.torch.aimdo_to_tensor(m._v, device_to)
                     allocated_size += v_weight_size
 
                 else:
@@ -1557,7 +1556,6 @@ def setup_param(self, m, n, param_key):
                         weight_size = geometry.numel() * geometry.element_size()
                         if vbar is not None and not hasattr(weight, "_v"):
                             weight._v = vbar.alloc(weight_size)
-                            weight._v_tensor = comfy_aimdo.torch.aimdo_to_tensor(weight._v, device_to)
                             weight._model_dtype = model_dtype
                         allocated_size += weight_size
                     vbar.set_watermark_limit(allocated_size)

comfy/ops.py

@@ -83,14 +83,18 @@ def cast_to_input(weight, input, non_blocking=False, copy=True):
 def cast_bias_weight_with_vbar(s, dtype, device, bias_dtype, non_blocking, compute_dtype):
     offload_stream = None
     xfer_dest = None
-    cast_geometry = comfy.memory_management.tensors_to_geometries([ s.weight, s.bias ])
 
     signature = comfy_aimdo.model_vbar.vbar_fault(s._v)
-    if signature is not None:
-        xfer_dest = s._v_tensor
     resident = comfy_aimdo.model_vbar.vbar_signature_compare(signature, s._v_signature)
+    if signature is not None:
+        if resident:
+            weight = s._v_weight
+            bias = s._v_bias
+        else:
+            xfer_dest = comfy_aimdo.torch.aimdo_to_tensor(s._v, device)
 
     if not resident:
+        cast_geometry = comfy.memory_management.tensors_to_geometries([ s.weight, s.bias ])
         cast_dest = None
 
         xfer_source = [ s.weight, s.bias ]
@@ -140,9 +144,13 @@ def cast_bias_weight_with_vbar(s, dtype, device, bias_dtype, non_blocking, compu
                     post_cast.copy_(pre_cast)
             xfer_dest = cast_dest
 
-    params = comfy.memory_management.interpret_gathered_like(cast_geometry, xfer_dest)
-    weight = params[0]
-    bias = params[1]
+        params = comfy.memory_management.interpret_gathered_like(cast_geometry, xfer_dest)
+        weight = params[0]
+        bias = params[1]
+        if signature is not None:
+            s._v_weight = weight
+            s._v_bias = bias
+        s._v_signature=signature
 
     def post_cast(s, param_key, x, dtype, resident, update_weight):
         lowvram_fn = getattr(s, param_key + "_lowvram_function", None)
@@ -182,7 +190,6 @@ def to_dequant(tensor, dtype):
     weight = post_cast(s, "weight", weight, dtype, resident, update_weight)
     if s.bias is not None:
         bias = post_cast(s, "bias", bias, bias_dtype, resident, update_weight)
-    s._v_signature=signature
 
     #FIXME: weird offload return protocol
     return weight, bias, (offload_stream, device if signature is not None else None, None)