GitSeonjeap
English

Sign in

Reading needs no sign-in. Sign in to write posts or comments, or to recommend and report, in the community.

By signing in you agree to the notices below. Community rules · Privacy notice

Keep browsing without signing in

← News

Local AI

Ollama v0.40.16-rc0: pplx-decider added, API behavior fixed

Posted As of

Ollama pre-release v0.40.16-rc0 adds Perplexity's 27B decider model, and changes how the OpenAI-compatible endpoint handles default sampler values. The stable release may differ from these notes.

What happened

Based on the official Ollama GitHub release notes.

  • pplx-decider, Perplexity's 27B (official) decider model fine-tuned from Qwen3.8-27B, has been added.
  • The OpenAI-compatible endpoint no longer overrides the sampler defaults included in the model. If temperature or top_p is not sent, the model's own values are applied.
  • A failure when continuing a conversation while switching between thinking and non-thinking models in the Responses API has been fixed. Ordering errors and duplicate events in client tool call streaming have also been fixed.
  • Cloud web search responses in /api/chat are now interpreted correctly, including when they are compressed.

Our take

Users who run Ollama through the OpenAI-compatible API should check their results, since they may change when sampling parameters are not sent. This is rc0, so the content may change in the stable v0.40.16 release. I have not verified pplx-decider's benchmarks or license.

Sources and related posts

More in Local AI

  1. llama.cpp b12200, Qualcomm OpenCL quantization kernel improvements (github.com)

    The llama.cpp b12200 release contains one change that reworks OpenCL q8_0 and small-batch MoE processing. No performance figures were published.

    Local AI| |1 sources |Our take included

  2. Moonshot Releases Open-Weight Model Kimi K3.1 Pro (huggingface.co)

    Moonshot AI has uploaded the weights of Kimi K3.1 Pro, a multimodal agent model, to Hugging Face. The parameter count is listed as 2.8T (official) and the model is under a dedicated license.

    Local AI| |1 sources |Our take included

  3. llama.cpp b12180 adds text-only Kimi-K3.1-Pro support… Vulkan detects ReBAR (github.com)

    On November 6, llama.cpp build b12180 added text-only Kimi-K3.1-Pro support. The same day, the Vulkan backend began detecting whether ReBAR is supported.

    Local AI| |1 sources

All news