D5: the author's one-line Ollama command, 2026-10-23 (UTC+8), Ollama 0.40.8.
$ ollama run hf.co/erikkannike/qwen3-4b-deslop-v4-GGUF:Q4_K_M "Say hi."
08:56:35
pulling manifest ... Error: max retries exceeded: Get "https://us.aws.cdn.hf.co/[signed URL removed]": redirect target not allowed: us.aws.cdn.hf.co resolves to non-public 198.18.0.1

real	1m5.071s
user	0m0.187s
sys	0m0.132s
exit=1

Cause: our box's DNS answers every outside name with 198.18.0.1 (proxy fake-IP); Ollama refuses download redirects to non-public addresses. Same failure as 'ollama pull' in today's Ollama test, so it is our network, not this model.

Fallback (documented Ollama GGUF import): both GGUF files downloaded with curl from Hugging Face (about 34 s each, 2,497,279,552 and 2,497,281,120 bytes; sha256 matches Hugging Face LFS ids, see 01-vendor-claims.txt), then:
$ cat Modelfile.deslop
FROM /workspace/tt/gguf/qwen3-4b-deslop-v4-Q4_K_M.gguf
$ cat Modelfile.base
FROM /workspace/tt/gguf/Qwen3-4B-Instruct-2507-Q4_K_M.gguf
$ ollama create deslop-local -f Modelfile.deslop; ollama create qwen3-base-local -f Modelfile.base  -> success, success
$ ollama show deslop-local -> architecture qwen3, 4.0B params, Q4_K_M, template source gguf_chat_template