Use min/max dynamic shape bounds for forward inputs/outputs

Backfill constant metadata method values in config.json from .pte

fix(v0.9.0): split text vs VL into <size>/ and vl_<size>/; colocate tokenizers per-size; drop Qwen3.5-* legacy dirs

Add stub root config.json for HF download counter

Add spec-compliant config.json files

Add spec-compliant config.json files

Restructure to MODEL_SPEC.md convention

Upload qwen3_5_vl_2b_xnnpack_8da4w.pte

Upload qwen3_5_vl_0_8b_xnnpack_8da4w.pte

Update README.md

Upload 6 files

Delete qwen3_5_2b_xnnpack_8da4w.pte

Delete qwen3_5_0_8b_xnnpack_8da4w.pte

Upload 2 files

initial commit