You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
* fix the `forward_return_lse` function in `BatchPrefillWithRaggedKVCache` class ([#337](https://github.com/flashinfer-ai/flashinfer/pull/337))
9
+
* fix the scheduler behavior of large page size ([#333](https://github.com/flashinfer-ai/flashinfer/pull/333))
10
+
11
+
### Features
12
+
13
+
* customize `logits_soft_cap` value ([#339](https://github.com/flashinfer-ai/flashinfer/issues/339)) ([a2498f5](https://github.com/flashinfer-ai/flashinfer/commit/a2498f511b354ce049bda6be320a24b73c719be3))
14
+
15
+
16
+
### Performance Improvements
17
+
18
+
* change minimal `kv_chunk_size` back to 128 ([#329](https://github.com/flashinfer-ai/flashinfer/issues/329)) ([f237f5f](https://github.com/flashinfer-ai/flashinfer/commit/f237f5f80199e2c433fcca750713c6e774693b58))
19
+
* more options for kv tile size ([#336](https://github.com/flashinfer-ai/flashinfer/issues/336)) ([bf2a6c7](https://github.com/flashinfer-ai/flashinfer/commit/bf2a6c7c05a82e0ee0ea04381d04b84327355b69))
0 commit comments