ggerganov's picture
ggml : adapt AMX to tensor->grad removal (llama/0)
8a67e9f