llama-server detection issue on windows build because of .exe suffix

[CleyFaye]

Describe the bug Attempting to start tabbyml with the windows distribution fails because the ".exe" suffix is added after detection of the llama-server binary path. The detection successfuly find the good file ("llama-server.exe") but tries to run "llama-server.exe.exe".

I:\tabbyml>dir
 Le volume dans le lecteur I s’appelle Slow NVME
 Le numéro de série du volume est 7498-EBCD

 Répertoire de I:\tabbyml

10/06/2024  09:27    <DIR>          .
10/06/2024  09:27    <DIR>          ..
31/10/2023  12:29        97 644 544 cublas64_12.dll
31/10/2023  12:29       551 793 152 cublasLt64_12.dll
31/10/2023  12:17           550 912 cudart64_12.dll
06/06/2024  05:40        39 277 568 llama-server.exe
31/10/2023  12:29           329 728 nvblas64_12.dll
04/03/2024  03:21                99 run.bat
06/06/2024  10:03                92 run2.bat
18/03/2024  10:31               143 scheduler.bat
06/06/2024  05:40        84 540 416 tabby.exe
24/05/2024  09:45        99 186 688 tabby_x86_64-windows-msvc-cuda122.exe
04/03/2024  02:17        78 586 368 tabby_x86_64-windows-msvc-cuda122_nightly.exe
              11 fichier(s)      951 909 710 octets
               2 Rép(s)  428 742 520 832 octets libres

I:\tabbyml>tabby serve --model StarCoder2-7B --device cuda
←[31mThe application panicked (crashed).←[0m
Message:  ←[36mFailed to start llama-server with command Command { std: "I:\\tabbyml\\llama-server.exe.exe" "-m" "C:\\Users\\Cley Faye\\.tabby\\models\\TabbyML\\Nomic-Embed-Text\\ggml/model.gguf" "--cont-batching" "--port" "30888" "-np" "1" "--log-disable" "--ctx-size" "4096" "-ngl" "9999" "--embedding" "--ubatch-size" "4096", kill_on_drop: true }: Le fichier spécifié est introuvable. (os error 2)←[0m
Location: ←[35mcrates\llama-cpp-server\src\supervisor.rs←[0m:←[35m74←[0m

(error is "file not found" in french)

Information about your version Tabby version 0.12 This is run on Windows 10 22H2

Information about your GPU

+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 552.12                 Driver Version: 552.12         CUDA Version: 12.4     |
|-----------------------------------------+------------------------+----------------------+
| GPU  Name                     TCC/WDDM  | Bus-Id          Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |           Memory-Usage | GPU-Util  Compute M. |
|                                         |                        |               MIG M. |
|=========================================+========================+======================|
|   0  NVIDIA GeForce RTX 3080 Ti   WDDM  |   00000000:2D:00.0  On |                  N/A |
|  0%   49C    P8             35W /  370W |     556MiB /  12288MiB |     10%      Default |
|                                         |                        |                  N/A |
+-----------------------------------------+------------------------+----------------------+

Additional context I added the ".exe" suffixes manually to the files from the 0.12 release, otherwise windows won't run them (this build issue is already fixed in 6263b2a495a240cd6c26c78028c0a853e0fb5397 I think)

It is possible I missed a parameter to specify by hand where llama-server.exe is located, although I didn't see something related to that in the code at first glance.

[CleyFaye]

Note that having both "llama-server" (no suffix) and a copy named "llama-server.exe" (with suffix) actually works, as a temporary workaround.

I suppose the detection grabs the first one then adds the suffix.

[wsxiaoys]

Fixing in https://github.com/TabbyML/tabby/issues/2358

shall be released soon with 0.12.1

[CleyFaye]

Great.

I really need to improve my searching skills for closed issues; sorry for the noise.

[wsxiaoys]

No problem - please check https://github.com/TabbyML/tabby/releases/tag/v0.12.1-rc.0 to see if it fixes the issue for you