- optimize(all): use torch.inference_mode instead of with torch.no_grad()
- fix(core): missing params passing to generate
- chore(logging): use warning instead of warn for deprecation
- optimize(gpt): stream triggering timing
- feat(stream): add test for issue #521