stik/qwen3.8-27b-archive
Watch
1
0
Fork
You've already forked qwen3.8-27b-archive
 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions
Provenance for a local Qwen3.8-27B mirror: checksums, sources, fetch recipe. Weights not served here.
1 commit 1 branch 0 tags 3 KiB
  • Python 100%
Find a file
Exact
Repository files (latest commit first)
Filename Latest commit message Latest commit date
stik 8e8f534a5c Qwen3.8-27B archive provenance: checksums, sources, fetch recipe
2026-08-17 21:50:29 +00:00
fetch.py Qwen3.8-27B archive provenance: checksums, sources, fetch recipe 2026-08-17 21:50:29 +00:00
README.md Qwen3.8-27B archive provenance: checksums, sources, fetch recipe 2026-08-17 21:50:29 +00:00
SHA256SUMS Qwen3.8-27B archive provenance: checksums, sources, fetch recipe 2026-08-17 21:50:29 +00:00

README.md

Qwen3.8-27B — local archive provenance

A record of a local mirror of Qwen3.8-27B (Apache-2.0), kept against upstream churn or takedown. This repo holds provenance only — checksums, sources, and the fetch recipe. The weights themselves are not served here; grab them from the upstream sources below.

What was archived

Artifact Source (Hugging Face) Approx size
Full weights, BF16 safetensors Qwen/Qwen3.8-27B ~55.6 GB
GGUF Q4_K_M ggml-org/Qwen3.8-27B-GGUF ~17 GB
GGUF Q8_0 ggml-org/Qwen3.8-27B-GGUF ~29 GB

GGUF quants are the llama.cpp project's own (ggml-org), not a third-party re-upload.

Verify

SHA256SUMS in this repo pins every archived file. After downloading from the sources above:

sha256sum -c SHA256SUMS

Reproduce the fetch

fetch.py is the exact script used (Python + huggingface_hub.snapshot_download). It pulls the full weights and the two GGUF quants, then writes the checksum manifest.

Notes

  • License: Apache-2.0 — redistribution permitted; this archive exists for availability, not because the upstream is going anywhere you can't reach.
  • Why provenance-only: 100 GB of weights do not belong in a git repo, and re-serving them from a small host is slower than the upstream CDN for everyone. The hashes let anyone confirm a copy pulled from Hugging Face is byte-identical to what was archived here.
  • Released: 2026-08-14 (Apache-2.0, 27B multimodal, 262K native context).