Skip to content

mp: use look-ahead actuals for stream offload VRAM calculation (fixes unwanted TE full offload) #11096

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
comfyanonymous merged 1 commit into from
Dec 4, 2025
Merged

mp: use look-ahead actuals for stream offload VRAM calculation (fixes unwanted TE full offload) #11096

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
12 changes: 9 additions & 3 deletions comfy/model_patcher.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -699,12 +699,12 @@ def load(self, device_to=None, lowvram_model_memory=0, force_patch_weights=False
offloaded = []
offload_buffer = 0
loading.sort(reverse=True)
for x in loading:
for i, x in enumerate(loading):
module_offload_mem, module_mem, n, m, params = x

lowvram_weight = False

potential_offload = max(offload_buffer, module_offload_mem + (comfy.model_management.NUM_STREAMS * module_mem))
potential_offload = max(offload_buffer, module_offload_mem + sum([ x1[1] for x1 in loading[i+1:i+1+comfy.model_management.NUM_STREAMS]]))
lowvram_fits = mem_counter + module_mem + potential_offload < lowvram_model_memory

weight_key = "{}.weight".format(n)
Expand Down Expand Up @@ -876,14 +876,18 @@ def partially_unload(self, device_to, memory_to_free=0, force_patch_weights=Fals
patch_counter = 0
unload_list = self._load_list()
unload_list.sort()

offload_buffer = self.model.model_offload_buffer_memory
if len(unload_list) > 0:
NS = comfy.model_management.NUM_STREAMS
offload_weight_factor = [ min(offload_buffer / (NS + 1), unload_list[0][1]) ] * NS

for unload in unload_list:
if memory_to_free + offload_buffer - self.model.model_offload_buffer_memory < memory_freed:
break
module_offload_mem, module_mem, n, m, params = unload

potential_offload = module_offload_mem + (comfy.model_management.NUM_STREAMS * module_mem)
potential_offload = module_offload_mem + sum(offload_weight_factor)

lowvram_possible = hasattr(m, "comfy_cast_weights")
if hasattr(m, "comfy_patched_weights") and m.comfy_patched_weights == True:
Expand Down Expand Up @@ -935,6 +939,8 @@ def partially_unload(self, device_to, memory_to_free=0, force_patch_weights=Fals
m.comfy_patched_weights = False
memory_freed += module_mem
offload_buffer = max(offload_buffer, potential_offload)
offload_weight_factor.append(module_mem)
offload_weight_factor.pop(0)
logging.debug("freed {}".format(n))

for param in params:
Expand Down