LLaVA vs. Inkling

LLaVA, which stands for Large Language-and-Vision Assistant, is an innovative multimodal model that integrates a vision encoder with the Vicuna language model, facilitating a deeper comprehension of visual and textual data. Through its end-to-end training approach, LLaVA demonstrates impressive conversational skills akin to other advanced multimodal models like GPT-4. Notably, LLaVA-1.5 has achieved state-of-the-art outcomes across 11 benchmarks by utilizing publicly available data and completing its training in approximately one day on a single 8-A100 node, surpassing methods reliant on extensive datasets. The development of this model included creating a multimodal instruction-following dataset, generated using a language-focused variant of GPT-4. This dataset encompasses 158,000 unique language-image instruction-following instances, which include dialogues, detailed descriptions, and complex reasoning tasks. Such a rich dataset has been instrumental in enabling LLaVA to efficiently tackle a wide array of vision and language-related tasks. Ultimately, LLaVA not only improves interactions between visual and textual elements but also establishes a new standard for multimodal artificial intelligence applications. Its innovative architecture paves the way for future advancements in the integration of different modalities.

Inkling is an open-weights multimodal AI model from Thinking Machines built to support customization, agentic workflows, coding, reasoning, vision, audio, and enterprise AI use cases. The model is a Mixture-of-Experts transformer with 975 billion total parameters, 41 billion active parameters, 256 routed experts per MoE layer, and six routed experts active per token. It supports context windows up to 1 million tokens and was pretrained on 45 trillion tokens across text, images, audio, and video. Inkling is designed as a broad foundation model rather than a narrowly optimized benchmark model, giving it balanced capabilities across reasoning, coding, factuality, instruction following, vision, audio, tool use, and safety. Its controllable thinking effort lets developers adjust how much computation and generated reasoning the model uses, helping teams balance quality, latency, and cost for different production needs. The model can run agentic coding tasks, use tools, create web apps, generate polished multi-page artifacts, reason over long contexts, and work through iterative refinement loops. For multimodal tasks, Inkling can process images, answer questions about visual content, transcribe and reason over audio, follow spoken instructions, and combine visual reasoning with code-based tools such as Python. Thinking Machines trained Inkling for calibration, instruction following, factual reliability, refusal behavior, and safety across multiple modalities, including evaluations for dangerous capabilities and human-AI threat vectors. Inkling is available on Tinker for fine-tuning, with 64K and 256K context options, an Inkling Playground for testing, cookbook recipes, and support for multimodal post-training workflows. Its full weights are available on Hugging Face, and deployment support is available through APIs and infrastructure partners such as TogetherAI, Fireworks, Modal, Databricks, Baseten, SGLang, vLLM, llama.cpp, and transformers.