Cog

A bug fix release to properly support models that use CUDA via the queueing API. See 573 for details.

Changelog
* 9e6ebe2 Bump pillow from 9.0.0 to 9.0.1
* c88b60c Use multiprocessing spawn method to support CUDA
* 75d0fbe update notebook docs to mention installing jupyterlab (563)

Hightlights

Cog now has an experimental ⚡️ [built-in queue worker](https://github.com/replicate/cog/blob/1cde0ae054f119ea850bf8bce49b3ed64196f691/docs/redis.md) ⚡️ that can process predictions from a Redis queue. See [docs/redis.md](https://github.com/replicate/cog/blob/1cde0ae054f119ea850bf8bce49b3ed64196f691/docs/redis.md).

This release also includes lots of bugfixes and improvements to the documentation.

Breaking changes
* 311ffc0 The `success` status has been change to `succeeded` in the HTTP API and throughout Cog

Changelog
* 1cde0ae Add experimental notice to queue API
* 3e1447b Add logs to queueing API responses
* e010d1e Allow version to be specified with make install
* 5d82ba0 Document that file handling is subject to change
* 0f7128f Document the Redis queue API
* 1ef7bbc Don't expect prediction ID from queueing API request
* 5e265ab Fix curl command and add real image
* 0defd8b Fix enums being passed to predict() when using `choices`
* bacaf22 Fix getting started guide
* 290d987 Include output and logs with failed responses
* 90c78e7 Remove input wrapping for queueing API
* af91725 Remove redis_log_capture
* 9b69a58 Rename inputs to input for queueing API requests
* 4f2271c Rename output key from "value" to "output"
* 311ffc0 Rename success status to succeeded
* 55875b0 Rework predictor subprocess to be long running
* 4d3bff2 Support complex outputs from predictors
* 5e1824e Use an Enum for whether output is a generator
* 5c7201b fix notebook docs

Changelog
* 8af71f6 Bump gotest.tools/gotestsum from 1.7.0 to 1.8.0 (543)
* 684c735 Enable file and numpy encode_json() tests (537)
* 04bd68a Only allow one of python_packages or python_requirements to be set
* 9fdba11 Recursively encode tuple outputs (553)
* 7510bfd Update Discord link (551)
* 2d86d20 Update companies doing this sort of thing

Changelog
* 32865a1 Add EXPOSE to Dockerfile (517)
* f1ca5cc Add Python 3.10 support for PyTorch
* 4591e1c Add documentation to explain how to deploy models (526)
* e05b0c1 Bump github.com/docker/docker
* bb64680 Bump github.com/getkin/kin-openapi from 0.92.0 to 0.93.0 (512)
* da71b5d Bump github.com/getkin/kin-openapi from 0.93.0 to 0.94.0
* 9ad0a58 Define some nouns
* e7abca8 Document HTTP API
* 3e1e608 Fetch latest torch versions from PyPi
* d1426c0 Fix output documentation
* 45f351a Install pyenv from github (525)
* 6724526 Make it clear what is incomplete (528)
* 26c82a0 More forgiving parser to deal with --extra-index-url
* d73af5c Pick latest patch version for CUDA
* 0528f3c Relax generate compatibility to handle ROCm
* dadab64 Rename run.cog.cog_version label to run.cog.version (532)
* e855d3d Revert "Update compatibility matrices (504)" (530)
* cb707f7 Switch to cog.run labels
* b057e19 Update compatibility matrices
* 488203e Update compatibility matrices (504)
* dd4d325 add -p flag to`cog run` to publish ports
* 9c8d6a2 document how to use notebooks with cog (524)

8c296dc Check if remote image exists before pulling in predict
c06a01d Debug statement in ImageInspect
063d0c0 Fix formatting
adf3295 Make remote models use GPUs in cog predict, if model requires GPU
76c3712 Relax torchvision version check
3741f20 Use r8.im in docs
290627d feat: add new command `cog init` (279)
37b570c minor tweak to function names in example
9346ce7 note that build runs before first prediction
f0a9fc7 update cog.yaml reference doc
c7e038a use curl instead of wget to download weights (271)

18e8fa4 Capture log output from generator
bcb9b44 Create extensions.json
e984c84 Document not downloading weights in setup()
b048b49 Set timeouts for brpop tests
78df85d [wip] Yield progressive output
7017729 clarify installation steps for Mac M1 users
4f2c573 clean up the wheel
13d874d document release process (236)
0ca463e make fmt
1ab99b2 refine M1 copy
7180bcf update http responder to handle generator functions
fdc2804 use sudo to install