mirror of
https://github.com/ollama/ollama.git
synced 2026-07-11 02:04:06 +00:00
64737330a4
The nvidia_fp32 config for (576, 512) head sizes had nbatch_fa=32, which caused zero-sized arrays when computing array dimensions: nbatch_fa / (np * warp_size) = 32 / (2 * 32) = 0 This resulted in CUDA compilation failures on CUDA 12 (Windows and Linux arm64): - "static assertion failed with nbatch_fa % (np*warp_size) != 0" - "the size of an array must be greater than zero" Fix by changing nbatch_fa from 32 to 64 for all (576, 512) configs in the nvidia_fp32 function, matching the nvidia_fp16 and AMD configs. |
||
|---|---|---|
| .. | ||
| bert | ||
| deepseek2 | ||
| deepseekocr | ||
| gemma2 | ||
| gemma3 | ||
| gemma3n | ||
| glm4moelite | ||
| gptoss | ||
| lfm2 | ||
| llama | ||
| llama4 | ||
| mistral3 | ||
| mllama | ||
| nomicbert | ||
| olmo3 | ||
| qwen2 | ||
| qwen3 | ||
| qwen3vl | ||
| qwen25vl | ||
| models.go | ||