README.md

metadata

title: README
emoji: 🚀
colorFrom: blue
colorTo: green
sdk: static
pinned: false

OnlyTextLLMs

OnlyTextLLMs provides text-only variants of state-of-the-art open-weights models.

Modern frontier models are predominantly released as native multimodal architectures: for deployments strictly constrained to natural language processing, retaining vision and audio encoders introduces unnecessary memory overhead and configuration complexity.

Focus Areas

• Pruned Checkpoints: Releasing model weights with vision encoders, audio modules, and associated projection layers surgically removed.
• Standardized Architectures: Converting complex multimodal configurations into standard AutoModelForCausalLM structures for seamless integration.
• Inference Optimization: Delivering leaner memory footprints to maximize VRAM availability for KV caching and context scaling.

---

All repositories maintain clear attribution and adhere to the licensing terms of the original model authors.