Announcing Tinker
13 hours ago
- #AI
- #Machine Learning
- #API
- Launch of Tinker, a flexible API for fine-tuning language models.
- Enables researchers to experiment with models by controlling algorithms and data.
- Supports fine-tuning of various open-weight models, including large ones like Qwen-235B-A22B.
- Managed service handling scheduling, resource allocation, and failure recovery.
- Uses LoRA for cost-efficient shared compute between training runs.
- Provides low-level API primitives and an open-source Tinker Cookbook for post-training methods.
- Already used by groups at Princeton, Stanford, Berkeley, and Redwood Research.
- Currently in private beta for researchers and developers.
- Free to start with usage-based pricing coming soon.