Releases: MalevichAI/malevich
v0.4.1
v0.3.1
Full Changelog: v0.3.0...v0.3.1
v0.3.0
What's Changed
- Commands/dev by @RamPrin in #42
- Remove gql from space interpreter and ability to invoke Space with SpaceOps inited outside #43 by @zakhardavydov in #44
Full Changelog: v0.2.6...v0.3.0
Contributors
Assets 2
v0.2.6
What's Changed
- app_cfg with class part, get_object by @pogrebnoijak in #27
- add has_object by @pogrebnoijak in #32
- Dev/unstable by @RamPrin in #33
- Dev/unstable by @RamPrin in #34
- Commands/docs by @RamPrin in #35
- Commands/dev by @RamPrin in #36
- added manifest show command by @RamPrin in #38
- Dev/unstable by @RamPrin in #40
- Fix: Windows temp files operations by @teexone in #41
Full Changelog: v0.2.4...v0.2.6
Contributors
Assets 2
v0.2.4: `has_object` for Square
add: has_object (#32)
v0.2.3: `get_object` Square interface
Full Changelog: v0.0.2...v0.2.3
v0.2.2
v0.0.2: CLI interface with image installers
app_cfg with class part, get_object (#27) (#30) * as_object, update as_df OBJ * app_cfg with class part, get_object --------- Co-authored-by: Andrew Pogrebnoi <pogrebnoijak@users.noreply.github.com>
v0.2.1
New
- Documentation. Explore https://malevichai.github.io/malevich/ for rich explanations about concepts used in Malevich, examples of use, and structured API references.
- CLI: Run
malevich newto create a new app and jump into writing code for apps faster.
Updates:
Results:
Working with results obtained by running flows on Malevich Core became more user-friendly. Use methods like get_df(), get_dfs(), get_binary(), get_binary_dir() to obtain results in desired format skipping manual mapping or type checking. Results come in strict and well-defined schema
Square
Now composite assets are converted to dataframes as a list of paths allowing you to straightly use paths in the table instead of resolving the type of assets. Also, you can now produce assets in the app by using context.as_object() function.
Batching
Couple of Core operation are now performing faster via Core Batcher interface.
New Registry
We changed our AWS registry ID, so malevich install will only be valid from this version
v0.2: Assets, Runners and many performance improvements
New:
Assets
Assets are local files, that can be uploaded and used directly in flows. You can pass them between apps, that support assets as input, upload them from your local system, and download them back from Core by returning them in flow
from malevich import asset, flow
from malevich.openai import speech_to_text
@flow()
def wonderful_flow():
audio = asset.file('speech.mp3')
return speech_to_text(audio)Runners
Changing data from run to run required a lot of boilerplate code. To enhance user experience and allow running a large amount of tasks from code with minimal effort, we introduce Runners:
from malevich import flow, collection
from malevich.runners.async_core import AsyncCoreRunner
from malevich.openai import prompt_completion
@flow()
def ask():
data = collection('question')
return prompt_completion(data, config=...)
runner = AsyncCoreRunner(task=ask, core_auth=(..., ...))
for q in question_bank:
runner.run(question=q, callback=print)
runner.wait()They automatically understand points, where you can inject data (e.g. collection with the name "question" in this case), and provides a neat and fast way of running and fetching results.
Minor Changes
Besides runners and assets, we improved the robustness and speed of our Core Interpretation, new features in Malevich Square including PyTorch Serialization and many others.