0.0.7

What's Changed

• Docker: Cuda11.8, make dependencies optional by @michaelfeil in #33
• Breaking changes: New install groups pip install infinity-emb[server,logging,onnx-gpu-runtime]
• add onnx-gpu by @michaelfeil in #26

Full Changelog: 0.0.6...0.0.7

0.0.6

What's Changed

• add python api by @michaelfeil in #32

Full Changelog: 0.0.5...0.0.6

0.0.5

What's Changed

• Docker image multi by @michaelfeil in #24
• patch missing event -> 200ms to 7ms inference at Batch size 1

Full Changelog: 0.0.4...0.0.5

0.0.4

What's Changed

PRs:

• Fastembed v2 by @michaelfeil in #21 :

Issues:
Closes #5 ONNX Support via https://github.com/qdrant/fastembed/
Closes #22 making pytorch and optional dependency

tl,dr

• fastembed as backend besides ct2 or torch
• v1/models returns "backend"
• makes torch an optional dependency
• calculates "min" sleep time dynamically on startup _> slightly optimized.
• default model is now "BAAI/bge-small-en-v1.5"

Full Changelog: 0.0.3...0.0.4

0.0.3

What's Changed

• add Flash-Attention+ optimum-BetterTransformers by @michaelfeil in #20
• Improve real-time / sleep strategy, async await for queues and result futures - reducing latency a bit by @michaelfeil in #12
• add better FIFO queueing strategy - your requests now have a upper bound how long they queue by @michaelfeil in #19

Docs:

• Docs: Update README.md by @michaelfeil in #8
• Update description. Update pyproject.toml by @michaelfeil in #9
• Refactor model dir by @michaelfeil in #10
• Update README.md by @michaelfeil in #14
• Update README.md by @michaelfeil in #15

Full Changelog: 0.0.2rc0...0.0.3

0.0.2

What's Changed

• Docs: Update README.md by @michaelfeil in #8
• Update description. Update pyproject.toml by @michaelfeil in #9
• Refactor model dir by @michaelfeil in #10
• Improve real-time / sleep strategy, async await for queues and result futures by @michaelfeil in #12

Full Changelog: 0.0.1...0.0.2rc0

0.0.1

Initial release of Infinity

0.0.1-dev3

What's Changed

• startup msg, log handling, import by @michaelfeil in #4
• update CI to release pypi by @michaelfeil in #7

Full Changelog: 0.0.1-dev2...0.0.1-dev3

0.0.1-dev2 - Speedups

adds new dependency (orjson) for faster response serialization - 300%
uses torch.inference_mode() and delayed moving to CPU - 10%
adds uvicorn[standard] - slightly faster 2-5%?
Updates readme

#2

0.0.1-dev1

This is a release for testing the CI of Infinity.