Comfy-Org / Comfy-Org/ComfyUI

int8 always load from disk

Open
#14,708 0 comments 0 reactions 0 assignees View on GitHub
Potential Bug
Dominant language
Python
Stars
133k
Forks
15.7k
Avg merge
1d 7h
Merged PRs (30d)
158

Description

bernini int8 , start with -disable-pinned-memory --reserve-vram 0.5. ram is cleared after executing,very strange

python 313tor212cu130

### Debug Logs

```powershell
[INFO] got prompt
[INFO] Model WanTEModel prepared for dynamic VRAM loading. 6419MB Staged. 0 patches attached. Force pre-loaded 73 weights: 488 KB.
[INFO] 0 models unloaded.
[INFO] Model WanVAE prepared for dynamic VRAM loading. 241MB Staged. 0 patches attached. Force pre-loaded 60 weights: 61 KB.
[INFO] 0 models unloaded.
[INFO] Model WanVAE prepared for dynamic VRAM loading. 241MB Staged. 0 patches attached. Force pre-loaded 60 weights: 61 KB.
[INFO] 0 models unloaded.
[INFO] Model WanVAE prepared for dynamic VRAM loading. 241MB Staged. 0 patches attached. Force pre-loaded 60 weights: 61 KB.
[INFO] 0 models unloaded.
[INFO] Model WanVAE prepared for dynamic VRAM loading. 241MB Staged. 0 patches attached. Force pre-loaded 60 weights: 61 KB.
[INFO] 0 models unloaded.
[INFO] Model WAN21 prepared for dynamic VRAM loading. 13858MB Staged. 893 patches attached. Force pre-loaded 160 weights: 1603 KB.
100%|████████████████████████████████████████████████████████████████████████████████████| 2/2 [00:34<00:00, 17.41s/it]
[INFO] Model WAN21 prepared for dynamic VRAM loading. 13858MB Staged. 893 patches attached. Force pre-loaded 160 weights: 1603 KB.
100%|████████████████████████████████████████████████████████████████████████████████████| 2/2 [00:34<00:00, 17.45s/it]
[INFO] Model WanVAE prepared for dynamic VRAM loading. 241MB Staged. 0 patches attached. Force pre-loaded 60 weights: 61 KB.
[INFO] Prompt executed in 93.43 seconds
```

Contributor guide

Open the contributing guide

Research direction

Reproduce the report with the bernini int8 model, -disable-pinned-memory, and --reserve-vram 0.5, using the provided debug-log pattern as the baseline. Trace the model-loading path responsible for the repeated dynamic-VRAM messages and confirm whether int8 weights are loaded from disk and whether RAM is released after execution; done means the behavior is explained and a focused regression check covers it.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
machine-learning, performance
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Quiet
Clarity
Needs clarification
Newbie friendliness
38/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.