Kimi K2 Thinking on serverless, GitHub evaluator templates and RFT step endpoints

Improved Evaluator Creation Experience

The evaluator creation workflow has been significantly enhanced with GitHub template integration. You can now:

• Fork evaluator templates directly from GitHub repositories
• Browse and preview templates before using them
• Create evaluators with a streamlined save dialog
• View evaluators in a new sortable and paginated table

MLOps & Observability Integrations

New documentation for integrating Fireworks with MLOps and observability tools:

• Weights & Biases (W\&B) integration for experiment tracking during fine-tuning
• MLflow integration for model management and experiment logging

✨ New Models

• Kimi K2 Thinking is now available in the Model Library
• KAT Dev 32B is now available in the Model Library
• KAT Dev 72B Exp is now available in the Model Library

☁️ Serverless

• Kimi K2 Thinking is now available on serverless

📚 New REST API Endpoints

New REST API endpoints are now available for managing Reinforcement Fine-Tuning Steps and deployments:

• Create Reinforcement Fine-Tuning Step
• List Reinforcement Fine-Tuning Steps
• Get Reinforcement Fine-Tuning Step
• Delete Reinforcement Fine-Tuning Step
• Scale Deployment
• List Deployment Shape Versions
• Get Deployment Shape Version
• Get Dataset Download Endpoint

[accordion] Bug Fixes & Minor Improvements

• Deployment Region Selector: Added GPU accelerator hints to the region selector, with Global set as default for optimal availability (Web App)
• Preference Fine-Tuning (DPO): Added to the Fine-Tuning page for training models with human preference data (Web App)
• Redeem Credits: Credit code redemption is now available to all users from the Billing page (Web App)
• Model Library Search: Improved fuzzy search with hybrid matching for better model discovery (Web App)
• Cogito Models: Added Cogito namespace to the Model Library for easier discovery (Web App)
• Custom Model Editing: You can now edit display name and description inline on custom model detail pages (Web App)
• Loss Curve Charts: Fixed an issue where loss curves were not updating in real-time during fine-tuning jobs (Web App)
• Deployment Shapes: Fixed deployment shape selection for fine-tuned models (PEFT and live-merge) (Web App)
• Usage Charts: Fixed replica calculation in multi-series usage charts (Web App)
• Session Management: Removed auto-logout on inactivity for improved user experience (Web App)
• Onboarding: Updated onboarding survey with improved profile and questionnaire flow (Web App)
• Fine-Tuning Form: Max context length now defaults to and is capped by the selected base model's context length (Web App)
• Secrets for Evaluators: Added documentation for using secrets in evaluators to securely call external services (Docs)
• Region Selection: Deprecated regions are now filtered from deployment options (Web App)
• Playground: Embedding and reranker models are now filtered from playground model selection (Web App)
• LoRA Rank: Updated valid LoRA rank range to 4-32 in documentation (Docs)
• SFT Documentation: Added documentation for batch size, learning rate warmup, and gradient accumulation settings (Docs)
• Direct Routing: Added OpenAI SDK code examples for direct routing (Docs)
• Recommended Models: Updated model recommendations with migration guidance from Claude, GPT, and Gemini (Docs)