tlg aa7a160118 fix: proper VRAM cleanup on model unload + CUDA alloc config ...

- Force gc.collect() before torch.cuda.empty_cache() to ensure all
  model references are released
- Set PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True in container

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

2026-04-05 17:59:23 +02:00

..

llmux

fix: proper VRAM cleanup on model unload + CUDA alloc config

2026-04-05 17:59:23 +02:00