Release Notes
v0.3.107
Autonomous Fine-tuning
- Added the ability to download fine-tuned models weights.
- Added the ability to download model weights from an AWS s3 url.
- Improvements to ML service.
v0.3.102
UI Improvements
- Upgraded the UI for the autonomous fine-tuning agent.
- You can now set a default job name.
- New tooltips in the autonomous fine-tuning agent.
- New button to copy the format of the json and jsonl.
- You can now copy paste to insert a dataset.
Autonomous Fine-tuning
- Added the ability to delete a fine-tuning job.
- You can now download the default dataset.
- Update default datasets
v0.3.101
- Added support for gemma 2, qwen2.5 and codellama models for fine-tuning and inference.
- Improved fine-tuning evaluation.
- Improvements to autonomous fine-tuning.
- Added system prompt for fine-tuned models.
- Improvements to fine-tuning start and completion emails.
- Improvements to login page responsiveness.
v0.3.100
- Improvements to data augmentation.
- The Gym is now Autonomous Fine-tuning.
v0.3.94
- Improvements to integrations messaging.
- Added support for Deepseek-r1-distill-70b model.
- Added support for Deepseek-r1 model.
v0.3.90
Finetuning
- Removed finetuning for Groq models.
- Added fine-tuning evaluation.
- Improved fine-tuning evaluation metrics.
- Improved fine-tuning Model deployment.
- Improved Docker Composed for fine-tuning.
Integrations
- UI updates for the integrations page.
- Fixed multiple integration connections.
- Improved integration deletion UX.
- Added Notion integration.
- Added Gmail integration.
- Added Github integration.
- You can now activate integrations through the Launchpad.
RAG
- Added the new bring your own Quadrant Vector Database feature.
v0.3.89
- Internal model evaluation.
- Added superuser topbar.
- Show images in Traces.
- Deprecated Groq “tool-use” models.
v0.3.87
Added support for the following new models:
- llama 3.3 70b - Powered by Groq
- claude-3.5-sonnet-v2
- claude-3.5-haiku
We’ve also added some new UI updates:
- New Navigation bar and sidebar. Now it’s easier to navigate the platform.
- Updated user login view
- Update privacy policy and terms links
Other additions:
- When a model is deleted, an email is sent to the user.
v0.3.80
- Refactored checkpointing into a scheduler
- Removed embedding model from repo update form
- Improved frontend tooling
- Support for Llama 70B full and lora finetuning
v0.3.72
Improvements
- Dark Mode Only: We’ve removed the light mode option, focusing on a more streamlined and cohesive dark mode experience.
- Datasource Improvements: We’ve applied various enhancements and fixes to datasource functionality, improving overall data integration and processing.
Features
- Document Deletion: You can now delete incomplete documents, allowing for more flexible content management.
- Pending Invites Management: We’ve introduced a new feature to handle pending invites, allowing users to manage their invitations effectively, including the ability to cancel pending invitations.
Release Notes
v0.3.107
Autonomous Fine-tuning
- Added the ability to download fine-tuned models weights.
- Added the ability to download model weights from an AWS s3 url.
- Improvements to ML service.
v0.3.102
UI Improvements
- Upgraded the UI for the autonomous fine-tuning agent.
- You can now set a default job name.
- New tooltips in the autonomous fine-tuning agent.
- New button to copy the format of the json and jsonl.
- You can now copy paste to insert a dataset.
Autonomous Fine-tuning
- Added the ability to delete a fine-tuning job.
- You can now download the default dataset.
- Update default datasets
v0.3.101
- Added support for gemma 2, qwen2.5 and codellama models for fine-tuning and inference.
- Improved fine-tuning evaluation.
- Improvements to autonomous fine-tuning.
- Added system prompt for fine-tuned models.
- Improvements to fine-tuning start and completion emails.
- Improvements to login page responsiveness.
v0.3.100
- Improvements to data augmentation.
- The Gym is now Autonomous Fine-tuning.
v0.3.94
- Improvements to integrations messaging.
- Added support for Deepseek-r1-distill-70b model.
- Added support for Deepseek-r1 model.
v0.3.90
Finetuning
- Removed finetuning for Groq models.
- Added fine-tuning evaluation.
- Improved fine-tuning evaluation metrics.
- Improved fine-tuning Model deployment.
- Improved Docker Composed for fine-tuning.
Integrations
- UI updates for the integrations page.
- Fixed multiple integration connections.
- Improved integration deletion UX.
- Added Notion integration.
- Added Gmail integration.
- Added Github integration.
- You can now activate integrations through the Launchpad.
RAG
- Added the new bring your own Quadrant Vector Database feature.
v0.3.89
- Internal model evaluation.
- Added superuser topbar.
- Show images in Traces.
- Deprecated Groq “tool-use” models.
v0.3.87
Added support for the following new models:
- llama 3.3 70b - Powered by Groq
- claude-3.5-sonnet-v2
- claude-3.5-haiku
We’ve also added some new UI updates:
- New Navigation bar and sidebar. Now it’s easier to navigate the platform.
- Updated user login view
- Update privacy policy and terms links
Other additions:
- When a model is deleted, an email is sent to the user.
v0.3.80
- Refactored checkpointing into a scheduler
- Removed embedding model from repo update form
- Improved frontend tooling
- Support for Llama 70B full and lora finetuning
v0.3.72
Improvements
- Dark Mode Only: We’ve removed the light mode option, focusing on a more streamlined and cohesive dark mode experience.
- Datasource Improvements: We’ve applied various enhancements and fixes to datasource functionality, improving overall data integration and processing.
Features
- Document Deletion: You can now delete incomplete documents, allowing for more flexible content management.
- Pending Invites Management: We’ve introduced a new feature to handle pending invites, allowing users to manage their invitations effectively, including the ability to cancel pending invitations.