Barry Nelson

Here is some of the rather persistent code that insists on having a vnc.html and loading a web server... ``` # Find vnc.html if [ -n "${WEB}" ]; then if...

Proposed patch... ``` --- noVNC-1.5.0/utils/novnc_proxy 2025-02-07 11:38:00.355963553 -0500 +++ noVNC-1.5.0/utils/novnc_proxy 2025-02-07 12:22:28.702594234 -0500 @@ -109,4 +109,16 @@ esac done +if [ -z "$WEB" ] +then + WEB="`pwd`" + if [...

This adds the ability to set --web NONE to disable the web server, otherwise the web folder defaults to the folder where the software is installed as it does now....

Now I just need to figure out (if I can) how to restrict access to a specific list of source IPs. For now I am just binding the server to...

I have already developed a patch to add the option not to start the web server to the novnc_proxy script and posted it here. As I did so I realized...

I tried CUDA 12.8 and CUDA 12.9, both gave this error using GCC 11. Are you maybe using a newer version of GCC? ``` [ 31%] Building CUDA object ml/backend/ggml/ggml/src/ggml-cuda/CMakeFiles/ggml-cuda.dir/mean.cu.o...

I am retrying with GCC 14 now.

Nope, same problem. ``` /usr/local/cuda/include/cuda/std/__cmath/traits.h(360): error: the global scope has no "isgreater" {return ::isgreater((type) __x, (type) __y);} ^ /usr/local/cuda/include/cuda/std/__cmath/traits.h(383): error: the global scope has no "isgreaterequal" {return ::isgreaterequal((type) __x, (type)...

GCC 13 same issue... ``` [ 26%] Building CUDA object ml/backend/ggml/ggml/src/ggml-cuda/CMakeFiles/ggml-cuda.dir/mean.cu.o nvcc warning : Support for offline compilation for architectures prior to '_75' will be removed in a future release...

Ok, well, something else must be different... Last CUDA toolkit version that works to build Ollama 0.12.10 on my system is 12.6. Using GCC 11,13 or 14 SuSE Leap 15.6...