qwen3.6-27b-gguf (Enclave model volume)

Curated model volume for Enclave confidential inference: Qwen3.6-27B at unsloth Dynamic Q4_K_XL (~18 GB), a single-file GGUF bundled with the official tokenizer.json so the volume is self-contained.

Upstream unsloth/Qwen3.6-27B-GGUF ships 21 quantizations plus vision mmproj-* files and an imatrix (~370 GB total) and no tokenizer.json. This volume carries ONLY the one served quant plus the tokenizer (same rationale as qwen3.5-9b-gguf / qwen3.5-122b-gguf): a multi-gguf volume makes the host's preload pick ambiguous, and wrapping the whole repo wastes ~350 GB of attested volume for files that never serve. The mmproj files are omitted deliberately: llm-chat's ggml path is text-only and preloads a single LLM gguf.

Provenance

File	Upstream	Revision	sha256
Qwen3.6-27B-UD-Q4_K_XL.gguf	unsloth/Qwen3.6-27B-GGUF	82d411acf4a06cfb8d9b073a5211bf410bfc29bf	ff6941ded525b34eb159496762c29dd0ec6e71dc31b74d57e75d871a03eec259
tokenizer.json	Qwen/Qwen3.6-27B	6a9e13bd6fc8f0983b9b99948120bc37f49c13e9	5f9e4d4901a92b997e463c1f46055088b6cca5ca61a6522d1b9f64c4bb81cb42

The tokenizer is byte-identical to the one in qwen3.5-9b-gguf and qwen3.5-122b-gguf: Qwen3.6 kept the Qwen3.5 family tokenizer (same sha256, same vocab 248320, <|im_end|> 248046 / <|endoftext|> 248044).

The GGUF declares general.architecture = qwen35 (the hybrid SSM + full attention layout, full_attention_interval 4), the same arch the fleet's pinned llama.cpp already serves for the Qwen3.5 volumes.

LICENSE is the Apache-2.0 text; both upstreams are Apache-2.0.

Usage on Enclave

Wrapped as a Tinfoil Modelwrap volume (dm-verity; the root hash is part of the enclave measurement). Deployments attach it by name and the guest reads it at /models/<name>; the host preloads the GGUF as the wasi-nn ggml graph. As the volume's single gguf it needs no MODEL_VOLUMES third field and no model_file override, and with the tokenizer bundled, llm-chat's default tokenizer.json lookup works with no cross-volume configuration.