Change log

• Correctly map num_predict to max_tokens.
• Important: Set num_predict to a positive value, such as 1024.
• Parameters (except num_ctx) can be left empty to use the engine’s default values.
• The system message can be empty to use the model’s default.
• You can now specify the num_gpu parameter to set the number of layers loaded to the GPU for Ollama.

Download