(Staging PR) Model management

[allozaur]

No description provided.

[allozaur]

Few things to address, plus:

• Make sure we have still proper llama-swap compatibility
• Unify imports for $lib/utils and $lib/types for cleaner code
• Add proper documentation of the WebUI architecture and functionalities to address ggml-org#16256

[allozaur]

Move to updated README in tools/server docs

[ggerganov]

I observe the following issue:

• Start a server with: ./bin/llama-server --port 8033
• Load gpt-oss-20b from the model selector
• In new chat and input something (e.g. "test")
• After model responds, stop the model by clicking the red power off button in the model selector
• Now load it again and try to enter a second message
• I get this error:

I think there is some bug with the command construction because I can see the process is started but it has two --port arguments:

ps ax | grep llama-server

92161 s000  S+     0:01.60 ./bin/llama-server --port 62516 -hf ggml-org/gpt-oss-20b-GGUF --alias ggml-org/gpt-oss-20b-GGUF --port 62495

[ngxson]

@ggerganov It should be fixed in my last commit from ggml-org#17470

@allozaur probably we no longer have to publish a PR anymore, because I regularly git merge your branch into mine. Or maybe you can just push directly to my PR: ggml-org#17470

[allozaur]

@ggerganov It should be fixed in my last commit from ggml-org#17470

@allozaur probably we no longer have to publish a PR anymore, because I regularly git merge your branch into mine. Or maybe you can just push directly to my PR: ggml-org#17470

Yeah, I today in the morning saw u had merged my changes. I mainly wanted to keep this as a separate PR for the time of working out the architecture, but from now I think I will be good with just pushing to your branch.