diff --git a/.github/workflows/apple-perf.yml b/.github/workflows/apple-perf.yml index 98fbb7ff4ff..56fc67d1617 100644 --- a/.github/workflows/apple-perf.yml +++ b/.github/workflows/apple-perf.yml @@ -230,7 +230,7 @@ jobs: model.use_sdpa_with_kv_cache=true \ backend.xnnpack.enabled=true \ backend.xnnpack.extended_ops=true \ - base.preq_mode="8da4w_output_8da8w" \ + base.preq_mode="preq_8da4w_out_8da8w" \ base.preq_group_size=32 \ export.max_seq_length=2048 \ export.max_context_length=2048 \ @@ -256,7 +256,7 @@ jobs: base.params="${DOWNLOADED_PATH}/params.json" \ quantization.use_qat=true \ base.use_lora=16 \ - base.preq_mode="8da4w_output_8da8w" \ + base.preq_mode="preq_8da4w_out_8da8w" \ base.preq_group_size=32 \ base.preq_embedding_quantize=\'8,0\' \ model.use_sdpa_with_kv_cache=true \