Webis

Appearance

Troubleshooting Guide ​

This article lists common errors and solutions when using Webis.

Cannot Access API: 401 Unauthorized ​

Reasons:

  • No valid API Key provided, or server has authentication enabled.

Solutions:

  • Check api_keys.json in the project root directory and copy a valid key.
  • Add api_key parameter when calling the API, for example:
bash
curl -X POST http://localhost:8000/generate   -H "Content-Type: application/json"   -d '{"prompt": "hello", "api_key": "your-key"}'

Model Loading Failed: Insufficient GPU Memory ​

Error:

text
ValueError: Free memory on device (...) is less than desired GPU memory utilization (...)

Reason:

  • Remaining GPU memory is less than the memory limit required by the model.

Solution:

  • Set lower GPU_MEMORY_UTILIZATION:
bash
export GPU_MEMORY_UTILIZATION=0.6

Cannot Find libcudart.so Error ​

text
ImportError: libcudart.so.11.0: cannot open shared object file: No such file or directory

Reason:

  • CUDA dynamic library not installed or version incompatible.

Solution:

  • Install PyTorch with corresponding CUDA via conda:
bash
conda install pytorch-cuda=12.1 -c pytorch -c nvidia

Build Failure: C Compiler Not Found ​

text
RuntimeError: Failed to find C compiler.

Reason:

  • Missing compilation tools like gcc, Triton failed when compiling model kernels.

Solution:

bash
sudo apt update
sudo apt install build-essential

HuggingFace Model Download Failure (WSL Network Issue) ​

Symptoms:

  • Model download stuck or prompts ConnectionError, Failed to establish new connection

Solution:

  1. Run in Windows CMD:
cmd
ipconfig

  1. Find the machine's IPv4 address, for example 192.168.0.123

  2. Set proxy in WSL:

bash
export http_proxy=http://192.168.0.123:7890
export https_proxy=http://192.168.0.123:7890

No Model Output / API Returns Empty String ​

Possible Reasons:

  • prompt content is incomplete or lacks context.
  • max_tokens set too low, causing generation to be truncated.

Suggestions:

  • Appropriately increase the max_tokens parameter (e.g., 256 → 512)
  • Use clear prompts, for example:
text
Please extract contact name and phone number from the following HTML: <html>...</html>

Package Manager Lock Error (dpkg/apt) ​

Error:

text
Waiting for cache lock: Could not get lock /var/lib/dpkg/lock-frontend. It is held by process

Reason:

  • Another package management process (apt, apt-get, dpkg, or system update) is running
  • Previous installation was interrupted, leaving lock files

Solution:

  1. If a legitimate update is running, wait for it to complete
  2. If no other updates are present, check which process is holding the lock:
bash
ps aux | grep -i apt
  1. If necessary, delete lock files (use with caution):
bash
sudo rm /var/lib/apt/lists/lock
sudo rm /var/lib/dpkg/lock 
sudo rm /var/lib/dpkg/lock-frontend
sudo dpkg --configure -a

Copyright © 2025-2026 Webis Development Team