remove v75 bundle: not Plane-B-validated (Plane B load-fail (6GB > ~4.7GB DSP ceiling; W4 HTP-blocked))

Restore v75 QHexRT HNPU artifacts

Prune QHexRT HNPU bundle to runtime-minimum artifacts

Add HNPU arch tags to sanitized model card

Sanitize public model card metadata

Add generic chat block to v79 manifest (template-as-data) (#3)

v75 (S24/SM8650, soc 57): Qwen3.5-4B sharded-W8 decode bundle — Plane A greedy-correct, 3.8 tok/s (#2)

Add v79 W8 sharded bundle (QAIRT 2.45; device-validated S25 3/3 greedy-exact, 5.1 tok/s) (#1)

Scrub internal engineering notes from manifest _comment

Add root config.json to enable Hub download counting

Upload v81/tokenizer.json with huggingface_hub

Upload v81/qwen354b_embed_f16.bin with huggingface_hub

Upload v81/qwen354b_lmhead_f16.bin with huggingface_hub

Upload v81/qwen354b_decode_s1_w8.bin with huggingface_hub

Upload v81/qwen354b_decode_s0_w8.bin with huggingface_hub

Upload v81/qwen3.5-4b-1024.json with huggingface_hub

Upload README.md with huggingface_hub

initial commit