Scoutari

AI builder 的決策情報台。

瀏覽

  • 今日情報
  • Skill 榜單
  • AI 週報
  • 關於與編輯方針
  • RSS

主題

  • 模型發布
  • API 與價格
  • MCP 與 Skills
  • AI Coding
  • 中國模型
  • 開源模型
  • Agent
  • Infra / 成本
  • 值得試用
© 2026 Scoutari · scoutari.com摘要由 AI 輔助生成，並始終附上原文連結。
情報模型收藏週報設定
Thursday, July 9, 2026Scoutari
Log inSign up
Sign up
ModelsSkill RadarSavedPersonalizeSubmit sourceLog in
IntelModelsSkill RadarWeeklySavedPersonalizeSubmit source
Back to timeline

Wed, July 807:03Open SourceModel releasesOpen sourceAgents

Ollama v0.31.2: GPU Compatibility Improvements and Core Agent Support

View originalGet it on GitHub

Decision Brief

What changedOllama v0.31.2 enhances CUDA and ROCm GPU support, enables Flash Attention for CUDA CC 6.x, and removes legacy code and experimental features.
Why it mattersEnabling Flash Attention on CUDA CC 6.x GPUs and adding a fallback strategy for JetPack CUDA improve inference performance and compatibility on older architectures and edge devices.
Who should careOpen-source model users, Inference / infra teams
Affected stackOllamaLlamaNVIDIA
Builder actionUpgrade Ollama to v0.31.2
Source confidenceHigh · Official release / blog / repo

Key changes in Ollama v0.31.2 include: fixed CUDA toolkit detection and parallel processing; automatic fallback to standard CUDA when JetPack runtime is missing; enabled Flash Attention on CUDA CC 6.x GPUs to boost inference speed; removed deprecated ROCm devices and dead code; removed OLLAMA_EXPERIMENT=client2 server variable; updated llama.cpp and MLX backends to latest versions. Developers using older NVIDIA GPUs (e.g., GTX 1080, CC 6.x) will see significant inference speedups from Flash Attention. In NVIDIA Jetson environments, if JetPack runtime is absent, the system auto-switches to standard CUDA, preventing service interruption. Additionally, ROCm no longer supports legacy hardware; affected users should watch for future updates.

Summary basis: official / RSS sourceCompiled from the source scope noted above; the original remains authoritative.

Sources

  • Ollama（GitHub Releases）

    Local-model runtime releases: new supported models and serving features.

  • Ollama（GitHub Releases）

Related intel

  • Ollama v0.31.2: Agent Core Framework & GPU Flash Attention

    Ollama v0.31.2 introduces an Agent core framework and enables Flash Attention for CUDA CC 6.x GPUs, along with various fixes.

  • Ollama v0.31.2 Release: Agent Core Integration, CUDA Fallback, and More

    Ollama v0.31.2 includes Agent core integration, CUDA fallback, Flash Attention enablement, MLX rewrite, and various fixes.

  • Ollama v0.31.1 Boosts Gemma 4 Token Generation by Nearly 90% on Apple Silicon with Multi-Token Prediction

    Ollama v0.31.1 uses multi-token prediction to significantly accelerate Gemma 4 inference on Apple Silicon.

  • Porting Moebius 0.2B Image Restoration Model to Browser with Claude Code

    Author uses Claude Code to port the Moebius 0.2B image restoration model, originally dependent on PyTorch and CUDA, to run in browsers with WebGPU support.

留言

登入後即可留言，和其他 builder 交換實測心得。

還沒有留言，搶頭香。