Toshiki Kataoka

The `tbump` instruction on README fails by ``` Error: Current version string: (0.1.0.dev0) not found in {{python package name}}/_version.py` ``` because `_version.py` constructs the version string `__version__` from `version_info`.

`requests_cache` is not optional on retry

1

## Description `requests_cache` is unconditionally imported at `uncache_url`, which is called by `fetch_with_retries` on retry. ## Reproduce `pytest --check-links` on a document containing a broken link that returns retry-after header,...

[Bugfix] [Core] don't schedule prefill if freeing kv cache

9

FIX #5578 This PR makes the priority "running > swapped > waiting" strict. The previous code has chance to schedule new prefills even if preemption/swap happens or there's a swapped...

[Bug]: Mamba should return states in fp32

1

### Your current environment N/A ### 🐛 Describe the bug There's a difference in the precision of SSM states. - Authors' implementation uses `weight_type`, which is usually fp32: https://github.com/state-spaces/mamba/blob/v2.2.4/csrc/selective_scan/selective_scan.cpp#L313 -...