cafonez/Escha-W2-35B-A3B-ROCmFP2

Confirm reproducibility: clean main build matches dev build performance

Restore --spec-mtp-strict-qwen (merged upstream in PR #54, commit 41db2f9eb)

Fix: --spec-mtp-strict-qwen is not in public builds; add build provenance caveat

Add ROCmFP2 GGUF (13.1GB, Q2_0_ROCMFPX experts + MTP head)

Add model card: flags, benchmarks, provenance

initial commit