Kumo-Tabular weights unreachable behind restrictive egress proxies (Xet + plain HTTP both fail)
Maintainers usually reply within 1 day
@JGagacheva is already working on this.
Since Oct 12, 2026.
Assessment
- Difficulty
- 2/5
- Estimated time
- 1-3 hours
- Newbie friendliness
- 67/100
- Issue type
- Documentation
- Clarity
- Mostly clear
- Activity status
- Active
- Tech stack
- python
- Domain
- machine-learning
Research direction
Start with the Kumo-Tabular model card or README and check how users are told to download the weights. Document the reported proxy failure and the limits of the tested workarounds, including HF_HUB_DISABLE_XET=1. Done means users can find this limitation and understand which download paths were tested; publishing another distribution path would need maintainer input.
Written by the indexing model from the issue text.
Description
Problem
nvidia/Kumo-Tabular weights cannot be downloaded in environments with a restrictive egress proxy. This blocks all evaluation/reproduction of the model outside direct-connect networks. Tested 2026-10-12 from a cloud VM (Ubuntu, Python 3.12, huggingface_hub current).
Reproduction (5 routes, all failed)
- Xet (default
huggingface_hubpath):hf_hub_download("nvidia/Kumo-Tabular", "small/classifier.pt", revision="v1.0.1")→CAS Client Erroragainstus.aws.cdn.hf.co. The Xet protocol does not survive the proxy. HF_HUB_DISABLE_XET=1(documented proxy fallback): falls back to plain HTTP, thenhttpx2.ReadTimeout— the CDN stream stalls indefinitely.curl -Lon the/resolve/URL: 307 redirect, then 0 bytes for 60s+.hf-mirror.commirror +HF_ENDPOINT: only.incompletefiles, same stall.- ModelScope: model not present there.
Note the front door works (HEAD/redirects return fine); only sustained file streams blackhole. pip, GitHub API, and small-file downloads through the same proxy are fine, so this is specific to large model-file streams.
Why it matters
Anyone behind a corporate/cloud egress proxy — a large share of enterprise users — cannot obtain the weights at all. The published benchmarks are therefore irreproducible for them, and the documented HF_HUB_DISABLE_XET=1 workaround does not help when the plain-HTTP stream itself stalls.
Suggestions
- Document the proxy limitation and the tested workarounds (and their limits) in the model card / README.
- Consider publishing a non-Xet distribution path (plain-HTTP-only blob host) or checksummed split archives downloadable with resume support.
- If the Hub team owns the Xet-through-proxy failure, a pointer to the upstream issue would help.
Happy to provide full logs. Environment details on request.
- Dominant language
- Python
- Stars
- 316
- Forks
- 27
- Avg merge
- 1d 3h
- Merged PRs (30d)
- 157
Getting set up
- No Dockerfile or Docker Compose file
- No pull request template
- Read the contributing guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
More from NVIDIA/structured-data-models
-
Difficulty 5/5 Over a week Newbie friendliness 35/100
NVIDIA/structured-data-models#1071 · 1 reaction ·
Maintainers usually reply within 1 day
-
Difficulty 4/5 3-5 days Newbie friendliness 55/100
NVIDIA/structured-data-models#1030 · 6 reactions ·
Maintainers usually reply within 1 day
-
Difficulty 5/5 Over a week Newbie friendliness 32/100
NVIDIA/structured-data-models#1022 ·
Maintainers usually reply within 1 day
-
Difficulty 5/5 Over a week Newbie friendliness 42/100
NVIDIA/structured-data-models#947 · 1 reaction ·
Maintainers usually reply within 1 day
-
Add more ProcessorsOpen
Difficulty 4/5 3-5 days Newbie friendliness 35/100
NVIDIA/structured-data-models#336 · 1 comment ·
Maintainers usually reply within 1 day
All issues in NVIDIA/structured-data-models
Similar issues
-
python triage
Difficulty 2/5 1-3 hours Newbie friendliness 82/100
microsoft/semantic-kernel#14590 ·
Maintainers usually reply within 3 days
-
Difficulty 2/5 1-3 hours Newbie friendliness 72/100
Maintainers usually reply within 1 day
-
Difficulty 2/5 1-3 hours Newbie friendliness 85/100
pinterest/pymemcache#662 ·
Maintainers usually reply within 1 day
-
Difficulty 2/5 Half a day Newbie friendliness 62/100
gnocchixyz/gnocchi#1527 ·
Maintainers usually reply within 1 day
-
Difficulty 2/5 1-3 hours Newbie friendliness 64/100
tenstorrent/tt-model-manager#181 ·
Maintainers usually reply within 1 day