tensorchord/modelz-py

Python SDK and CLI for modelz.ai, which is a developer-first platform for prototyping and deploying machine learning models.

modelz-py's Issues

feat(py): Support Environment variable MODELZ_API_TOKEN

Bug: Modelz guide not work

Guide:

SDK
pip install modelz-py
You could use the following code to make a prediction request:

import modelz

# Or you could set the environment variable MODELZ_API_KEY
key = "xxx"

client = modelz.ModelzClient(key=key, project="text-xieydd-mosec-sd-1ooppa18x0pqaxuu.modelz.tech")
Check out [SDK documentation](https://docs.modelz.ai/sdk) for more details.

CLI
modelz inference --key xxx --deployment text-xieydd-mosec-sd-1ooppa18x0pqaxuu.modelz.tech --params prompt="cute cat"
Please add --serde if you want to use custom serialization. (e.g. --serde msgpack)

Check out [CLI documentation](https://docs.modelz.ai/cli) for more details.

Log:

 File "/usr/local/lib/python3.9/site-packages/httpx/_transports/default.py", line 77, in map_httpcore_exceptions
    raise mapped_exc(message) from exc
httpx.ConnectError: [SSL: CERTIFICATE_VERIFY_FAILED] certificate verify failed: Hostname mismatch, certificate is not valid for 'text-xieydd-mosec-sd-1ooppa18x0pqaxuu.modelz.tech.modelz.io'. (_ssl.c:1123)

modelz-py/modelz/env.py

Line 11 in db93745

self.host: str = "https://{}.modelz.io/"

feat(CICD): Add CICD for pypi

fix: Use the new domain logic

https://sd-w41s2qknzjo0fk3i.cloud.modelz.dev/

or

https://sd-w41s2qknzjo0fk3i.modelz.io

Bug: Update sdk document

https://docs.modelz.ai/sdk

$ python3 test-sd-mosec.py
Traceback (most recent call last):
  File "/Users/xieyuandong/Workspace/tencent/projects/tensorchord/modelz/test/test-sd-mosec.py", line 7, in <module>
    client = modelz.ModelzClient(key=key, deployment="text-xieydd-mosec-sd", serde="msgpack")
TypeError: __init__() got an unexpected keyword argument 'serde'

test: add CLI and client mock test

• serde
• CLI read-stdin/write-file
• save the response to local files
• config

feat: support multiple options

modelz inference --deployment imagebind-XXX --serde msgpack model=imagebind-text input="dog" input="cat" input="sheep"

could be parsed as:

input = ["dog", "cat", "sheep"]

example: https://click.palletsprojects.com/en/8.1.x/options/#multiple-options

feat(py): Support environment variable MODELZ_ENDPOINT for host

feat(CLI): Retry 3 times by default

This could avoid cold start failure. WDYT

@kemingy

feat: --host argument to choose any remote or local host

inf_parser.add_argument(
        "--host", help="host", default="https://{}.modelz.io/"
    )

client = ModelzClient(key=key, deployment=deployment, host=host)

The value could be one of:

• https://{}.xxx.yyy/ for formatting Deployment Key
• http://127.0.0.1:8000/ for no Deployment Key

Or maybe an environment variable MODELZ_API_HOST could be implemented?