Releases: RedisAI/aibench
Releases · RedisAI/aibench
Version 0.3.1
Features
- Added Tensorflow Lite benchmarking support (#56)
- Add per tick memory stats to redisai_vision runner. (#61)
- Re-use per-worker vision tensors as means for optimized vision benchmark (#59)
- Included a small script to process the JSON output and provide an CPU table with info for backend, main-thread, and background workers (#55)
- Vision multi endpoint benchmarks (#52)
🧰Maintenance
- Added missing data-vision and data-fraud make rules (#54)
- Included details on how to produce the results table after running the vision benchmark (#58)
Thanks to @koleini for the contributions!
Version 0.3.0
Changes
🚀Features
- Runtime metrics collection PR ( focusing on RedisAI benchmarks ) (#46)
- Retrieving futher runtime statistics on RedisAI vision benchmark (#44)
- Storing runtime info text output on the redisai vision benchmark (#43)
- Added json-output-file with overall inference stats. Improved overall vision testing script (testing autobatching, and tensor batching) (#40)
- Increased vision dataset size (#36)
- RedisAI vision binaries enhancements (#31)
- Added docs section regarding batching multiple inputs (images) to 4D batch tensor. Included also info on how to retrieve runtime stats (#30)
- Support for tensor batching on redisai vision benchmark (#29)
Bug Fixes
- Fixed large vision datasets generation (#39)
- Fixed missing iteration on vision data generator (#38)
- Fixed large vision datasets generation (#37)
- Fixed opencv install issue (#35)
- Fixed benchmark folders being wrongfully ignored (#34)
🧰Maintenance
- Splits all CI into 3 jobs ( build, dataset check for fraud, dataset check for vision ) (#47)
Version 0.2.0
Main Features
-
Included TensorFlow Serving on ARM supplemental guide for fraud benchmark @filipecosta90 (#22)
-
Financial benchmark refactor, go.mod across vendor dependencies and CI validation on both fraud and vision benchmarks @filipecosta90 (#24)
-
Update documentation to reflect all benchmarks (vision and fraud) install instructions @filipecosta90 (#23)
v0.1.1
This is a patch release for the benchmark version used for 1.0 benchmarks.
v0.1
- This version was used for RedisAI 1.0 results