Files
zeekay eeb464bb67 Rename the qwen4exp architecture to zen6
Every name the engine owns for the Qwen3.8-Flash-Next architecture is
now zen6: models::zen6 and models::quantized_zen6, Zen6Loader,
NormalLoaderType::Zen6 and its `--arch zen6` string,
GGUFArchitecture::Zen6, the zen6_moe fixtures, the zen6_* golden
scripts and LLM.md.

Upstream spellings are read, never written, where a checkpoint names
its architecture: the Hugging Face class
Qwen4ExpForConditionalGeneration in from_causal_lm_name, and the GGUF
general.architecture qwen4exp as a parse alias on
GGUFArchitecture::Zen6.  Content::from_readers moves such a file's
qwen4exp.* keys under zen6.*, so the lookups built from the enum's name
(block_count and head_count in the device mapper, block_count in the
pipeline) find them.

Recorded goldens are untracked: rename a local
tests/fixtures/qwen4exp{,_vectors}.safetensors to zen6{,_vectors}.
2026-09-28 14:09:56 -07:00
..