Files
Eric BuehlerandGitHub ee98ba196f Implement Imatrix for ISQ (#949)
* Implement imatrix for ISQ

* Implement imatrix quantization for q3k, q2k

* Fix build on cuda

* Implement imatrix quantization for q5k, q6k

* Utilize merged rev

* Fix typo

* Update docs

* Update docs

* Track stats for imatrix

* Implement internal imatrix quantization

* Implement internal imatrix quantization for all models but the moe ones

* Add for the moe models

* Simpler

* Handle clearing kv cache

* Add calibration data

* Add better calibration data

* Add example and docs

* Add example and docs
2024-11-30 06:55:43 -05:00
..
2024-11-30 06:55:43 -05:00