v1.9.1

Features:

• cpu embeddings model thenlper/gte-base/cpu
• manual upload weights for offline mode

v1.9.0

Features:

• new 3rdparty providers: Gemini, Grok (XAI), OpenAI o1
• offline mode: server now works with preloaded models without connection to huggingface

Minor fixes:

• caps reload on model list change fix
• removed deprecated models

v1.8.0

Refact.ai Self-hosted:

• CUDA and cuDNN Version Update: 11.8.0 -> 12.4.1
• New models: llama3.1, llama3.2 and qwen2.5/coder families.
• New providers support: 3rd party APIs for groq and cerebras.
• Support for multiline code completion models.

v1.7.0

Refact.ai Self-hosted:

• New models: last OpenAI models is now available in Docker.
• Tool usage support for 3rd party models: turn on 3rd party APIs to use latest features of Refact.
• Removed deprecated models.

v1.6.4

Refact.ai Self-hosted:

• Claude-3.5 Sonnet Support: New model from Anthropic is now available in Docker.

Refact.ai Enterprise:

• Llama3 vLLM Support: Added vLLM version of Llama-3-8B-Instruct for better performance.

v1.6.3

Refact.ai Self-hosted:

• Llama3 8k Context: llama3 models now support 8k context.
• Credentials Management: We added information about tokens and keys.
• Deprecated Models: The models starcoder, wizardlm, and llama2 are deprecated and will be removed in the next release.

Refact.ai Enterprise:

• Refact Model 4k Context: refact model now supports 4k context.

v1.6.2

Refact.ai Self-hosted

• Models Support: We've introduced support for gated models and the new llama3 model
• Even More Models: GPT4o and GPT4-turbo models are now available

Refact.ai Enterprise

• VLLM Speed Improvement: You are now able to experience faster processing times with our optimized VLLM
• VLLM LoRa-Less Mode: In cases where LoRa is not set up, VLLM will now operate 20% faster due to the new LoRa-less mode
• Empty Prompt and OOM Handling: We've addressed issues in VLLM that caused broken generations

v1.6.1

Context Switching Mechanism

We've implemented the context-switching mechanism, and it's available in our latest version of the VS Code plugin. Now, you can change the max context value for models depending on your needs — small context for less memory usage and faster operation or large context for deeper insights.

Model Deprecation

Our UI updates now flag models slated for removal. This ensures you're always working with the latest and most efficient models.

Factory Reset Fix

We've resolved issues with the factory reset process for when you need a fresh start.

v1.6.0

Refact.ai Self-hosted Updates:

Multiple Source Projects:

• Versatile Source Handling: In what used to be a single "sources" tab, you can now create multiple projects, each with its diverse sources.
• Fine-Tune on Demand: You can now fine-tune on your specific project.

Fine-Tune Enhancements:

• Unified Fine-Tuning Process: We've simplified the fine-tuning process by merging the filter and fine-tuning steps into one process.
• Multi-GPU Support: You can use multiple GPUs for faster fine-tuning!
• Simultaneous Fine-Tuning: Execute multiple fine-tuning processes concurrently to save time.
• Full Model Support: Load full model patched LoRA weights (note: requires substantial RAM).

Refact.ai Enterprise Updates:

• Customization Tab: Personalize system prompts and toolbox commands specifically for your team.
• Keycloak Integration: Secure user authentication with Keycloak, including a dedicated account page for users, ensuring both convenience and security. Check out the documentation for more information: https://docs.refact.ai/guides/keycloak/

v1.5.0

• Fine-tune Process Enhancements: We've made the fine-tuning process for starcoder models both faster and of higher quality with new default settings
• Fine-tune UI: The fine-tune setup has been moved to the Model Hosting tab for easier access
• Plugin Fine-tune Switching: VS Code and JetBrains plugins now support switching between fine-tuned models
• Chat Tab Redesign: The Chat tab has been temporarily hidden for a redesign and will be back in the next release

Compatibility Issues

• Plugin Support: Older versions of plugins will fall back to using the base model as they do not support the new fine-tuning capabilities. Make sure to update your plugin