* feat: add essentials_category to nodes and blueprints for Essentials tab
Add ESSENTIALS_CATEGORY or essentials_category to 12 node classes and all
36 blueprint JSONs. Update SubgraphEntry TypedDict and subgraph_manager to
extract and pass through the field.
Fixes COM-15221
Amp-Thread-ID: https://ampcode.com/threads/T-019c83de-f7ab-7779-a451-0ba5940b56a9
* fix: import NotRequired from typing_extensions for Python 3.10 compat
* refactor: keep only node class ESSENTIALS_CATEGORY, remove blueprint/subgraph changes
Frontend will own blueprint categorization separately.
* fix: remove essentials_category from CreateVideo (not in spec)
---------
Co-authored-by: guill <jacob.e.segal@gmail.com>
If a subclass BYO _load_from_state_dict and doesnt call the super() the
needed default init of these weights is missed and can lead to problems
for uninitialized weights.
This is an experimental WIP option that might not work in your workflow but
should lower memory usage if it does.
Currently only the VAE and the load image node will output in fp16 when
this option is turned on.
After a frontend update (e.g. nightly build), browsers could load
outdated cached index.html and JS/CSS chunks, causing dynamically
imported modules to fail with MIME type errors and vite:preloadError.
Hard refresh (Ctrl+Shift+R) was insufficient to fix the issue because
Cache-Control: no-cache still allows the browser to cache and
revalidate via ETags. aiohttp's FileResponse auto-generates ETags
based on file mtime+size, which may not change after pip reinstall,
so the browser gets 304 Not Modified and serves stale content.
Clearing ALL site data in DevTools did fix it, confirming the HTTP
cache was the root cause.
The fix changes:
- index.html: no-cache -> no-store, must-revalidate
- JS/CSS/JSON entry points: no-cache -> no-store
no-store instructs browsers to never cache these responses, ensuring
every page load fetches the current index.html with correct chunk
references. This is a small tradeoff (~5KB re-download per page load)
for guaranteed correctness after updates.
* Implement seek and read for pins
Source pins from an mmap is pad because its its a CPU->CPU copy that
attempts to fully buffer the same data twice. Instead, use seek and
read which avoids the mmap buffering while usually being a faster
read in the first place (avoiding mmap faulting etc).
* pinned_memory: Use Aimdo pinner
The aimdo pinner bypasses pytorches CPU allocator which can leak
windows commit charge.
* ops: bypass init() of weight for embedding layer
This similarly consumes large commit charge especially for TEs. It can
cause a permanement leaked commit charge which can destabilize on
systems close to the commit ceiling and generally confuses the RAM
stats.
* model_patcher: implement pinned memory counter
Implement a pinned memory counter for better accounting of what volume
of memory pins have.
* implement touch accounting
Implement accounting of touching mmapped tensors.
* mm+mp: add residency mmap getter
* utils: use the aimdo mmap to load sft files
* model_management: Implement tigher RAM pressure semantics
Implement a pressure release on entire MMAPs as windows does perform
faster when mmaps are unloaded and model loads free ramp into fully
unallocated RAM.
Make the concept of freeing for pins a completely separate concept.
Now that pins are loadable directly from original file and don' touch
the mmap, tighten the freeing budget to just the current loaded model
- what you have left over. This still over-frees pins, but its a lot
better than before.
So after the pins are freed with that algorithm, bounce entire MMAPs
to free RAM based on what the model needs, deducting off any known
resident-in-mmap tensors to the free quota to keep it as tight as
possible.
* comfy-aimdo 0.2.11
Comfy aimdo 0.2.11
* mm: Implement file_slice path for QT
* ruff
* ops: put meta-tensors in place to allow custom nodes to check geo
