SAGE-MM-Qwen2.5-VL-7B-SFT-GGUF

The SAGE-MM-Qwen2.5-VL-7B-SFT from allenai is a 7B-parameter vision-language model fine-tuned from Qwen/Qwen2.5-VL-7B-Instruct, serving as the core decision-maker in the SAGE (Smart Any-Horizon Agent) system for long video reasoning through a two-stage process: Stage-1 analyzes initial sampled frames and metadata to classify queries as single-turn (immediate answers) or multi-turn (tool-required), while Stage-2 iteratively generates JSON-formatted tool calls for web-search, speech transcription on timestamps, event grounding, video part extraction, and detailed visual analysis to build context progressively [attached_file:1 equivalent]. Designed to handle arbitrary-length videos beyond fixed horizons—like sports events, narratives, or complex timelines—it requires the SAGE GitHub runtime for tool parsing/execution and observation feedback, enabling robust Q&A via dynamic tool orchestration under Apache 2.0 for research/educational use per Ai2 guidelines, with GGUF quantizations for efficient deployment. This SFT variant powers the SAGE framework's superior performance on benchmarks like MINERVA, outperforming prior Qwen3-VL-4B baselines in extended video comprehension.

SAGE-MM-Qwen2.5-VL-7B-SFT [GGUF]

Quants Usage