AI
APM-Bench: Benchmarking Cross-session Persistent Memory for Egocentric Streaming Video Assistants
Publication: Hugging Face Daily Papers
Author: Not available in collected metadata
Source published: 28/09/2026, 20:00:00 UTC
Briefing created: 30/09/2026, 11:04:31 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 30/09/2026, 10:18:24 UTC.
Briefings
APM-Bench introduces a benchmark for evaluating persistent memory in streaming video assistants by reformulating real-world interactions as multi-session life trajectories. The dataset contains 549 sessions, 104 trajectories, and 2,719 candidates spanning objective and open-ended questions. Models are evaluated under various memory protocols to assess their ability to maintain reliable long-term recall, low overhead, and effective proactive assistance across interruptions. The evaluation reveals a clear utility-latency-storage trade-off, highlighting challenges in balancing memory efficiency with real-world utility.
Key points
- APM-Bench reformulates real-world streaming interactions into multi-session life trajectories with 549 sessions, 104 trajectories, and 2,719 candidates.
- Models are evaluated under diverse memory protocols to determine if they can maintain reliable long-term recall, low overhead, and effective proactive assistance across interruptions.
Context and limitations
The benchmark focuses on streaming video models and evaluates memory protocols under realistic streaming conditions, but specific performance metrics for each protocol are not provided in the evidence.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 30/09/2026, 11:04:31 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeie2l6niiic6n6nmma5e2efjkco4q6hppz6ib3jssrxlp7xmsnq5vm