Misc. bug: Remote download fails behind proxy with -hf flag

[yaoluxun]

Name and Version

bash-4.4$ ./llama-server --version
version: 8724 (b54cb2e3d)
built with GNU 13.3.1 for Linux x86_64

Operating systems

Linux

Which llama.cpp modules do you know to be affected?

llama-server

Command line

./llama-server -hf hugging-quants/Llama-3.2-1B-Instruct-Q8_0-GGUF -v

Problem description & steps to reproduce

1. Build

cmake -S . -B build \
  -DLLAMA_OPENSSL=ON

cmake --build build --config Release -j 

2. Behavior

bash-4.4$ ./llama-server -hf hugging-quants/Llama-3.2-1B-Instruct-Q8_0-GGUF -v 

common_download_file_single_online: no previous model file found ~/.cache/llama.cpp/hugging-quants_Llama-3.2-1B-Instruct-Q8_0-GGUF_preset.ini 
common_download_file_single_online: HEAD failed, status: -1 
no remote preset found, skipping

3. Troubleshooting

I ran

curl -I https://huggingface.co/
curl -I https://huggingface.co/hugging-quants/Llama-3.2-1B-Instruct-Q8_0-GGUF

Both succeed and return:

HTTP/1.0 200 Connection Established
HTTP/2 200

So the host can reach Hugging Face through the configured proxy.

4. Expected behavior

Remote download should work when the host has working HTTPS connectivity, or at minimum fail with a clearer error explaining why the HTTP request could not be completed.

I locally patched common/http.h to read HTTPS_PROXY and apply proxy settings to the underlying HTTP client, and that appears to be the right direction for a fix.

First Bad Commit

No response

Relevant log output

Logs

[oengelha]

I faced the same issue and can confirm patching http.h fixes the issue.

[polaco1782]

I can confirm this patch fixes it. I had run in the same issue, our corporate firewall has a proxy we have to manually set HTTP_PROXY and HTTPS_PROXY env variables. Without the patch llama-cli does not works with -hf parameter.

With the patch applied, it honored env variables and works just fine.

[arno4000]

is there any news regarding this? I used llama.cpp about five months back and then it worked behind a forward proxy. without this working, we can't use llama.cpp in a current version in our environment.