#vLLM

共 12 篇带此标签

12 articles Newest 2026-11-15
Archive · #vLLM 11 articles
Case Studies

A 4B Model Just Beat Postgres at Query Planning

QORL uses a 4B-parameter model to generate query plans end-to-end, beating PostgreSQL's cost-based optimizer by 81% on latency. The interesting part isn't the speedup, it's what it signals about replacing traditional database internals with small specialized models.

2026-09-17 Source 2026-09-16 6 min 135 views
LLMs

DeepSeek V4.1 Flash Hits HN Front Page: What 331 Points Tell Us

DeepSeek's new small model landed on Hacker News with 331 upvotes. By skipping the V4 milestone and going straight to Flash, the company is targeting edge deployment and long-context workloads rather than competing head-on with frontier models.

2026-09-10 Source 2026-09-10 6 min 135 views