Re-calibrate baseline_sqnr_db tags of 18 ggml gemm workloads
#12
by ArtificialRay7579 - opened
Updates only the baseline_sqnr_db tag in 18 workloads/gemm/gemm_ggml_*.jsonl files (q4_K x10, q5_K x2, q6_K x6; no other fields changed). Values were re-measured after the bf16 tensor-loader fix, which previously read real-activation workloads as integer bit patterns.
This is just for the support of #11 e2e rerun, don't merge #8 before merging this PR