Open Source
找到 11 篇关于此标签的文章
法国的 Linux 转型：为何主权代码是独立开发者的新护城河
法国弃用 Windows 的举措凸显了向开源主权战略转型的趋势。了解独立开发者如何利用 Linux 和 FOSS 降低成本并增强控制力。
Emotion Vector Steering in Qwen-2.5-3B: D318 Always Suppressive
A developer built an emotion vector steering pipeline on Qwen-2.5-3B, finding dimension 318 consistently suppresses emotional output.
MiniMax-M2.7 Open-Source Release Delayed to This Weekend
MiniMax delays M2.7 open-source release due to infrastructure work, now targeting this weekend.
Local AI Goes Mainstream When the Tooling Becomes Boring Infrastructure
A Reddit argument: local LLM adoption hinges on reliable tooling stacks, not benchmark gains, mirroring Docker's container revolution.
Gemma 4 31B Matches Gemini 2.5 Pro on Local Hardware Benchmarks
Community benchmarks show Gemma 4 31B achieving Gemini 2.5 Pro-level scores when run locally via llama.cpp harness.
开发者用 1900 年前文本训练 LLM，意外“重新发现”相对论
开发者 Michael Hla 仅使用 1900 年前的文本从零训练小型 LLM，模型在提示下部分推导出了量子力学与相对论的核心概念。
OpenCode Third-Party UI: When Official Tools Fail, What's Your Backup?
A JavaScript bug in OpenCode's official web UI caused hours of downtime, revealing the risk of single-point dependency on official tooling.
scan-for-secrets 0.1.1: Lightweight Secret Detection CLI Tool
Simon Willison releases scan-for-secrets 0.1.1 with improved escaping scheme docs and cleaner JSON detection logic.
Hermes Agent: Best Open-Source Local LLM Agent Framework in 2025
Nous Research's Hermes Agent offers per-model tool call parsers, Ollama/vLLM support, and MIT license at 22k stars.
Gemma-4-31B Multi-Agent Swarm Matches Gemini Pro and GPT-5 Benchmarks
A LocalLLaMA user built a Gemma-4-31B agent swarm achieving performance comparable to frontier closed models.
Qwen3.6-397B-A17B: First Open Model to Match Claude Sonnet in Real Use
Community testing finds Qwen3.6-397B-A17B matches Claude Sonnet reliability in real tasks, beating GLM-5.1 and Kimi-k2.5.