mirror of
https://github.com/ikawrakow/ik_llama.cpp.git
synced 2026-02-24 23:24:13 +00:00
Fix max nodes (again) (#1306)
This commit is contained in:
@@ -430,8 +430,8 @@ struct llama_model {
|
||||
if (arch == LLM_ARCH_QWEN3NEXT || arch == LLM_ARCH_QWEN35MOE) {
|
||||
return std::max<size_t>(n_tokens * 40, 32u * n_tensors);
|
||||
}
|
||||
return std::max<size_t>(1024, 8*n_tensors);
|
||||
//return 65536 * 2;
|
||||
//return std::max<size_t>(1024, 8*n_tensors);
|
||||
return 65536;
|
||||
}
|
||||
|
||||
bool has_tensor_overrides() const {
|
||||
|
||||
Reference in New Issue
Block a user