SOURCE-REVIEWED MODEL PROFILE
ALIVE-8B
Qwen3-VL-based streaming video assistant trained for monitoring, task coordination, feedback and silence.

Metadata reviewed 2026-11-21 · Original creator: ALIVE-Team
PARAMETERS8B class; about17.5GB BF16 checkpoint file
CONTEXT131,072-token web-demo default
LICENSEApache-2.0 (model-card declaration)
MODALITYVideo, images and text to text
What it does
- Streaming visual input
- Feedback-seeking interaction
- Public checkpoint
Local fit
Documented voice-enabled demo uses two CUDA GPUs. Main model on one GPU; Qwen ASR/TTS share a second. No12GB fit or speed test.
Configuration lists262,144 positions; working context and runtime memory differ.
Evidence and limits
Source and metadata reviewed; no inference or benchmark.
Source reviewed
License and source
Minimal model card declares Apache-2.0; no standalone LICENSE listed. Code has Apache2 and upstream notices; speech/dependency terms separate.
Read the original model card ↗
Reviewed source revision: 0acb034cd569ae673cb552783d1e853647de183d