Models · Updated 23 Oct, 07:42 pm IST
Ollama publishes v0.40.10-rc0 with Gemma 4 renderer and DFlash support
Why it matters for readers: It gives you a clearer view of how assistant content and tool calls can be ordered and how model internals are exposed for advanced decoding.
- A new gemma4-nemotron renderer was added that renders assistant content before its tool calls, and models opt in with RENDERER gemma4-nemotron while gemma4 auto-detection remains unchanged.1
- A chat template was added as a test fixture to check Jinja parity.1
- The mlx code now supports Gemma 4 targets with DFlash drafts and will load a draft-owned output head when required while preserving the existing target-head path for DFlash models.1
- The release exposes Gemma 4 hidden-state taps and raw logits to enable speculative decoding and reports safetensors target precision independently of quantized draft weights.1
Get a brief like this every morning
Uzha reads hundreds of sources and gives you the stories that matter for your work, with every source linked. Free.
Get started