-
Notifications
You must be signed in to change notification settings - Fork 929
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Unable to run TGI following the instructions on the readme #2058
Comments
Hey @xianbaoqian, could you share a bit about what is the exact problem you're seeing? I see two problems above, and I'm unsure of which one is affecting you, I see this one first:
Here it seems like you're calling I'm happy to help out, but it would be helpful if the formatting of the issue was easier to parse 🤗 thanks! |
Hi @LysandreJik the problem is that after conda activate I still can't use text-generation-launcher as the binary file can't be found: (base) root@C.11121520:$ conda activate text-generation-inference |
I updated the style (surrending it with a code block) to make it easier to read. The original format is indeed a bit misleading! |
Hey @xianbaoqian, thanks for clearing the style. There is definitely an issue in the code you run:
Here you're only creating a new conda environment and activating it. You named it If you took the example from this part of the README, it seems that you forgot this line: + BUILD_EXTENSIONS=True make install # Install repository and HF/transformer fork with CUDA kernels
text-generation-launcher --model-id mistralai/Mistral-7B-Instruct-v0.2 Beware: it's a long installation. Using TGI with the docker image is recommended as much faster. |
I had the same issue. Resolved by adding "$HOME/text-generation-inference/target/release-opt/" to PATH. On another note, cargo (or Rust) version 1.78.0 will not work as the Rust source files are written using inline const feature which is only available in cargo >= 1.79.0 . |
System Info
uname -a
Linux a3eb1d6a21b4 5.4.0-174-generic #193-Ubuntu SMP Thu Mar 7 14:29:28 UTC 2024 x86_64 x86_64 x86_64 GNU/Linux
cargo --version
cargo 1.78.0 (54d8815d0 2024-03-26)
nvidia-smi
Wed Jun 12 07:26:30 2024
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 550.67 Driver Version: 550.67 CUDA Version: 12.4 |
|-----------------------------------------+------------------------+----------------------+
| GPU Name Persistence-M | 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 4090 On | 00000000:81:00.0 Off | Off |
| 30% 28C P8 17W / 450W | 1MiB / 24564MiB | 0% Default |
| | | N/A |
+-----------------------------------------+------------------------+----------------------+
+-----------------------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=========================================================================================|
| No running processes found |
+-----------------------------------------------------------------------------------------+
Information
Tasks
Reproduction
Expected behavior
expects TGI to be running correctly without issues since the above is copied from installation section in the readme.md
The text was updated successfully, but these errors were encountered: