Ollama: Inkling on Mac Gains and Stream Reliability Fixes

MLX work adds Inkling 2 Mini, fuses Metal kernels for ~9% generation gains, and auto-sizes parallelism from memory. Parallel fixes tighten truncated streams, mid-stream errors, and runner queue pressure, plus preview installs and checksum verification.

Duration: PT2M36S

Episode overview

This episode is a short developer briefing from Ollama.

It explains recent repository work in plain language.

  • Show: Ollama
  • Published: 2026-11-21T14:01:51Z
  • Audio duration: PT2M36S

Transcript excerpt

This excerpt keeps the crawler page concise. Listen to the episode or use the RSS feed for the full update.

It's Saturday, November 21st, 2026, and this is your Ollama developer briefing.

The throughline today is Apple Silicon performance, plus a push to make partial results and busy servers fail visibly instead of silently.

First, MLX on Mac. PR 19799 landed support for Inkling 2 Mini, followed by follow-up tuning in PR 19806. PR 19808 fuses residual mixing — normalization, scoring and weighted sum — into a single Metal dispatch, reporting about a 9 percent generation speedup for single-user tests on M5 Max, and smaller prompt gains.…

Second, stream and scheduler reliability. Open PR 19791 would have the Go chat and generate clients surface an unexpected end-of-file error when a 200 response closes before completion, so partial answers aren't mistaken for full ones. Two related proposals, PRs 19795 and 19797, address the OpenAI-compatible…

Finally, install and hygiene. Merged PR 19801 adds a preview channel option to the install scripts, while open PR 19813 would verify Linux bundle checksums before extraction. Merged test work in PR 19742 steadies Pi catalog timeouts on macOS loopback.

What to watch: if you run on Mac, expect higher default concurrency and faster Inkling decode; if you…

Nearby episodes from Ollama

  1. Adaptive Memory on Macs, Tighter Server Security
  2. Tool Call Reliability and Runner Efficiency
  3. Runner Reuse, Decision Models, and Thinking UI Fixes
  4. MLX Concurrency and Memory Tuning
  5. GPU Load Fixes and Clearer Errors
  6. Weekly Recap - Parallel Inference, Tool Fidelity, and Launcher Polish
  7. Qwen Tool-Call Reliability Fix
  8. Tool Schema Fidelity, Desktop Launch, and Download Safety