v0.71.0: Split fails with "downloaded 17741 bytes but expected 288 bytes for model-package.json"

[takesan923]

Hi, I'm trying to use split inference with:

mesh-llm serve --model hf://meshllm/Qwen3-32B-UD-Q4_K_XL-layers --split

I'm getting this error on v0.71.0:

Failed to inspect model hf://meshllm/Qwen3-32B-UD-Q4_K_XL-layers:
download layer package manifest: download layer package file:
model-package.json: downloaded 17741 bytes but expected 288 bytes
for model-package.json

The model-package.json in the repo has skippy_abi_version: “0.1.22”
and is 17741 bytes. The layer-package was also requeued once
(queue.json shows: “requeued_reason”: "previous layer package repo
was incomplete").

Is this a known incompatibility between v0.71.0 and the current
layer-package format? Is there a workaround or a newer build that
supports skippy_abi_version 0.1.22?

---