Description
LiquidAI d1-omni-600M is a sophisticated decision model featuring 600 million parameters, built on the LFM2.5-Encoder-350M framework. This model is engineered to handle multimodal inputs, including text, images, and audio, in a single forward pass. It is particularly suited for applications requiring yes/no answers, option selections, or ratings, such as routing, moderation, classification, and visual inspection tasks.
Read More
The model's architecture includes a shared trunk and decision head, along with specialized encoders for vision and audio. The vision encoder is based on the SigLIP2 vision tower, while the audio encoder utilizes a 17-layer FastConformer. This design ensures that all modalities are processed using the same trunk weights, enhancing efficiency and consistency.
One of the key features of d1-omni-600M is its ability to return typed answers directly from the model's distribution over options, eliminating the need for output token generation or parsing. This capability is particularly beneficial for edge deployments where speed and resource efficiency are critical.
The model supports a context length of 16,384 tokens, accommodating text, image, and audio positions together. When images are included, the state and question text is limited to 896 tokens, as per training specifications. The model's vocabulary size is 65,536, allowing for a broad range of input processing.
Audio capabilities were specifically trained using interactions between an English speaker and an assistant, focusing on identifying utterance types, topics, and speaker intents. The model is not designed for chat applications and does not generate text, making it ideal for decision-making tasks.
LiquidAI d1-omni-600M is an open-source model, encouraging further research and development in the field of artificial intelligence. It is available for use with transformers version 5.15 or higher, and can be integrated into various pipelines requiring efficient decision-making capabilities.
- CategoryAI Models & LLMs
- Pricing ModelOpen source
- DeploymentCloud
- PlatformsWeb
LiquidAI d1-omni-600M Highlights
600M parameters
Multimodal input support
Single-pass decision making
Edge deployment optimized
Open source
Vision and audio encoders
No output token generation
Context length of 16,384 tokens
Getting Started with LiquidAI d1-omni-600M
Access model page
Authenticate
Configure model
Send prompts
LiquidAI d1-omni-600M's Use Cases
- Routing and Triage
- Moderation
- Intent Classification
- Voice Command Routing
- Visual Inspection
