Release Notes v1.2.0 🚀

Features 🌟

Distinguish Merged QKV for more precise and efficient handling of query, key, and value transformations.
Now LoRA can be applied to the MPT model.
Introduced support for Mixtral model checkpoint conversion.

Bug Fixes and Improvements 🐛🔨

AWQ bug during using gpt-j.
Addressed a critical issue where CUDA Out of Memory (OOM) errors occurred while using AWQ.
Minor update to phi model type.

friendli-client - Release 1.1.0 🚀

Published by kooyunmo 10 months ago

The features related to Friendli Dedicated Endpoints are temporarily removed from the client package. Please use periflow-client instead to use the features.

[!NOTE]
We are actively integrating the features in Friendli Dedicated Endpoints (previously known as PeriFlow Cloud) to Friendli Suite.
Please use periflow-client pacakge instead of friendli-client to use features for Friendli Dedicated Endpoints.
Those features will be supported with friendli-client very shortly.

CLI commands are updated.
- Commands related to the Friendli Dedicated Endpoints are removed temporarily.
- Commands for API calls to the serverless endpoints are added. Check the example usage below:

friendli api chat-completions create \
  -g "user Tell me how to make a delicious pancake" \
  -m llama-2-13b-chat

friendli-client - Release 1.0.0 🚀

Published by kooyunmo 10 months ago

🌟 Exciting Major Version Update: Introducing Friendli Suite! 🌈

We're thrilled to announce the official release of Friendli Suite, bringing a wealth of enhancements and features to your fingertips.
With this major update, we've given our package and GitHub repository a facelift, transitioning from periflow-client to the all-new and improved friendli-client.

Here's a rundown of the key changes:

🔄 CLI Command Prefix Update

The CLI command prefix has undergone a transformation! Say goodbye to the old pf and embrace the fresh friendli. Now, to sign in, simply use friendli login instead of the previous pf login.

🐍 Python SDK Breaking Change

In the Python SDK, we've introduced a breaking change that aligns its semantics with the OpenAI Python SDK v1. Fear not, as this update brings more consistency and compatibility. Dive into the details and explore examples on our documentation page.

import os
from friendli import Friendli

client = Friendli(api_key=os.environ.get("FRIENDLI_API_KEY"))

chat_completion = client.chat.completions.create(
    model="llama-2-13b-chat",
    messages=[
        {
            "role": "user",
            "content": "Tell me how to make a delicious pancake"
        }
    ],
    stream=False,
)

print(chat_completion.choices[0].message.content)

🚀 Upgrade now to unlock a world of possibilities with Friendli Suite! If you encounter any challenges or have questions, don't hesitate to reach out. Happy coding! 🤖💬

friendli-client - Release 0.1.13 🚀

Published by kooyunmo 11 months ago

Support checkpoints with safetensors format.

Full Changelog: https://github.com/friendliai/periflow-client/compare/0.1.12...0.1.13

friendli-client - Release 0.1.12 🚀

Published by kooyunmo 12 months ago

Now you can convert HuggingFace adapter models to PeriFlow-compliant format. Run pf convert-adapter --help to see details.

friendli-client - Release 0.1.11 🚀

Published by kooyunmo 12 months ago

Minor: Strict package version check is disabled.
The "offloading" option is added to the quantization config file. This option enables the GPU to CPU offloading to save the GPU memory usage.

friendli-client - Release 0.1.10 🚀