v6.2.5

What's Changed

• Fix bitrot by @jeremyfowers in #326

The Lemonade SDK has moved to https://github.com/lemonade-sdk/lemonade

Full Changelog: v6.2.4...v6.2.5

v6.2.4

What's Changed

• Release v6.2.4 with Lemonade relocation notice by @jeremyfowers in #325

The Lemonade SDK has moved to https://github.com/lemonade-sdk/lemonade

Full Changelog: v6.2.3...v6.2.4

v6.2.3

Summary of Changes

⚡Significant new Lemonade Server features (@danielholanda):

• Added OpenAI Responses API support
• Added Tool Calling support in Chat Completions API

📖Documentation Improvements

• A new Jupyter notebook under examples/lemonade/notebooks that provides an e2e tutorial for the Lemonade SDK tools (@jeremyfowers)
• Added GAIA and PEEL links to the Lemonade Server apps list, and linked videos for Continue and AI Toolkit (@vgodsoe)
• The Lemonade Server Models page now includes the full lemonade-server pull MODEL_NAME command for each model (@jeremyfowers)

🦗Bugs fixed:

• lemonade-install --ryzenai hybrid will detect whether you have a compatible PC and raise a helpful error message if you do not, which should eliminate the confusing unordered map errors people have been seeing (@ramkrishna2910)
• Improved the way lemonade-server CLI is added to PATH during GUI installation (@danielholanda, @jeremyfowers)
• Total time telemetry in Lemonade Server debug logging is now accurate (@jeremyfowers)
• Froze a variety of external packages (Hugging Face Hub, ONNX, ORT) to avoid bugs

🛠️Additional Changes:

• lemonade-server version now only displays the version number, to make it easier to parse (@kovtcharov)

New Contributors

@kovtcharov made their first contribution!

Full Changelog: v6.2.2...v6.2.3

v6.2.2

Server Improvements

v6.2.2 is focused on improvements for 🍋Lemonade Server users and app developers.

• LLMs can be downloaded after Lemonade Server installation using lemonade-server pull MODEL_NAME (@jeremyfowers)
  • Users can find a list of supported LLMs here.
  • Developers can programmatically access a list of supported LLMs here.
• New supported models:
  • 6 new AWQ OGA+CPU models, enabling performant and lightweight support for non-STX systems (@iswaryaalex)
  • Mistral-7B-v0.3-Instruct-Hybrid and Llama-3.1-8B-Instruct-Hybrid (@jeremyfowers)
• Stop any running Lemonade Server process using lemonade-server stop (@danielholanda)
• Lemonade Server works in offline mode (@ramkrishna2910)

Additional Improvements

• A new Lemonade Server integration guide, for using Mindcraft to add LLM-controlled NPCs to Minecraft (@itomek)
• Add response length to bench statistics and improve Quark installation process (@amd-pworfolk)

New Contributors

@itomek made their first contribution!

Full Changelog: v6.2.1...v6.2.2

v6.2.1

What's Changed

• Enable logprobs in OGA models, Hugging Face models, and Lemonade Server (@ramkrishna2910, @danielholanda)
• New Lemonade Server application integration guides:
  • LM Evaluation Harness, aka lm-eval (@ramkrishna2910)
  • wut-cli Linux terminal assistant (@jeremyfowers)
  • AnythingLLM GUI (@danielholanda)
• Major documentation improvements:
  • Overhaul for easier discoverability of server examples, contribution, and more (@vgodsoe)
  • Document lemonade.api for the first time in lemonade_api.md (@vgodsoe)
  • Lemonade Server's models are documented for the first time in server_models.md (@jeremyfowers)
• A variety of fixes for Lemonade Server (@danielholanda, @jeremyfowers)

Full Changelog: v6.2.0...v6.2.1

v6.2.0

What's Changed

• Ryzen AI NPU is available as a device backend in lemonade-install (@amd-pworfolk)
• Ryzen AI NPU and Hybrid execution can be installed into the same Python environment (@amd-pworfolk)
• Allow users to not create a desktop shortcut when using the installer (@danielholanda)
• Ability to select port on lemonade-server serve (@danielholanda)
• Add landing page to server's root URL (@danielholanda)
• Replace the HF/Torch CPU model in Lemonade Server with an OGA-CPU model (@jeremyfowers)
• Remove -m from Installer.nsi (@jeremyfowers) (in case that was causing #306)
• Build optimized OGA checkpoints for Ryzen AI NPU and Hybrid (@iswaryaalex)
• Add max_tokens and temperature parameters to Lemonade Server OpenAI endpoints (@danielholanda)
• Add checkpoint and recipe for each model on /api/v0/models endpoint (@danielholanda).

Changes to Experimental Lemonade Server Endpoints

• The experimental /load endpoint has been reworked (@jeremyfowers).
  • Load a model by its Lemonade name (e.g., Llama-3.2-1B-Instruct-Hybrid)
  • Load a model by its checkpoint and Lemonade Recipe (e.g., "checkpoint": "amd/Llama-3.2-3B-Instruct-awq-g128-int4-asym-fp16-onnx-hybrid", "recipe":"oga-hybrid")

For additional details, see load spec.

Full Changelog: v6.1.4...v6.2.0

v6.1.4

What's Changed

• Add chat templates to the llm-prompt tool (@amd-pworfolk)
• Replace Conda with embeddable Python in Lemonade Server installer (@jeremyfowers)
• Add --trust-remote-load option to oga-load tool (@amd-pworfolk)
• Fix a Ryzen AI specific bug related to context length (@jeremyfowers)
• Uplift Continue instructions to Continue v1.0 changes (@jeremyfowers)
• Ensure Lemonade Server shortcuts do not exit immediately when the server is already running (@danielholanda)

v6.1.3

What's Changed

• Fix default model selection on devices that do not support Hybrid (@danielholanda)
• Add a new lemonade-server CLI for starting server and checking status (@danielholanda)

Full Changelog: v6.1.1...v6.1.3

v6.1.1

What's Changed

• Upgrade Ryzen AI SW to version 1.4.0 (@amd-pworfolk, @jeremyfowers)
  • Add DeepSeek Hybrid models to Lemonade Server (@danielholanda)
  • Refactor the oga-load tool and oga.py (@ramkrishna2910)
• Documentation overhaul (@vgodsoe)
• New Lemonade Server demos:
  • CodeGPT (@vgodsoe)
  • Microsoft AI Toolkit (@danielholanda)
• Fixes:
  • Make sure that OGA models use their chat template in Lemonade Server (@danielholanda)
  • Lemonade API can load checkpoints from folders on disk using lemonade.api.from_pretrained() (@amd-pworfolk)

Full Changelog: v6.0.3...v6.1.1

v6.0.3

Breaking Changes

OpenAI-Compatible Server Model Selection

Lemonade's server now requires models to be downloaded at install time. Apps that use our installer in silent mode now have to specify which models to download. See docs/lemonade/server_integration.md for details.

Summary of Contributions

• Add guide on how to use Continue app with Lemonade Server (@jeremyfowers)
• Overhaul the lemonade help menu (@jeremyfowers)
• Stop importing tkml CLI in lemonade CLI (@jeremyfowers)
• Only show hybrid models when Hybrid is available (@danielholanda)
• Fix oga seed and avoid default params from being overwriten (@jeremyfowers)
• Improve Server Integration Documentation (@danielholanda)
• Add exception handler for server's generate thread (@jeremyfowers)
• Improve server logger in debug mode (@jeremyfowers)
• Added mmlu accuracy test command format (@vgodsoe)