China Upload Bottleneck Analysis

China Upload Bottleneck Analysis | Generated by AI

Home 2026.06

Test file cleaned up. Here’s the summary:

Smoke test results (China → HuggingFace):

  Direct:        ~190 KB/s (3.4MB in 17.9s)
  Via proxy 7890: ~157 KB/s (10MB in 65s, with long stalls)

  475MB estimated: ~42-50 minutes, likely to stall mid-transfer

Both routes are equally slow — the bottleneck is China’s international bandwidth, not the routing path. Proxy doesn’t help.

Your real options for model.safetensors:

Just run it overnight — direct upload, no proxy needed. It’ll be slow but if it doesn’t stall, ~50 min. hf upload does support resumable behavior for large files (multipart upload). If it stalls, rerun the same command — it picks up where it left off:
```
cd ~/projects/sec-edgar-gpt-124m
hf upload lzwjava/sec-edgar-gpt-124m model.safetensors model.safetensors
```
AMD Dev Cloud — but uploading 475MB from China→Atlanta (scp) faces the same wall. Only worth it if you can get the file there via GCS (which is also stuck).
Workstation (192.168.1.36) — also in China, same problem. Timed out on SSH.

Bottom line: Option 1 is your best bet. Kick it off before bed, let it grind through. If it stalls, rerun — hf upload re-uploads the file. Want me to start it in background?

Back Donate