Em-Garde-7B
Em-Garde-7B is a language model designed for low-latency streaming video understanding.
It is the IGPP Module of the Em-Garde framework, producing candidate proposals to be matched by the downstream embedding module.
Em-Garde is intended for use in systems that perform proactive streaming video understanding, such as video assistants, interactive agents, and robotics perception pipelines.
Model Description
- Model name: Em-Garde-7B
- Parameters: ~7 billion
- Architecture: Transformer-based causal language model
- Primary role: Proposal generation for streaming video reasoning
- Designed for: Low-latency and streaming-compatible inference
Training Pipeline
Em-Garde-7B is initialized from Qwen2.5-VL-7B-Instruct and finetuned through SFT and RL stages.
