Update

[ghstack-poisoned]

Author: jackzhxng

.github/workflows/android-perf.yml

@@ -228,7 +228,7 @@ jobs:
                       export.output_name="${OUT_ET_MODEL_NAME}.pte" \
                       model.use_kv_cache=true \
                       model.dtype_override=fp32 \
-                      base.preq_embedding_quantize='8,0' \
+                      base.preq_embedding_quantize=\'8,0\' \
                       quantization.use_spin_quant=native \
                       base.metadata='{"get_bos_id":128000,"get_eos_ids":[128009,128001]}'
                     ls -lh "${OUT_ET_MODEL_NAME}.pte"
@@ -249,7 +249,7 @@ jobs:
                       base.use_lora=16 \
                       base.preq_mode="8da4w_output_8da8w" \
                       base.preq_group_size=32 \
-                      base.preq_embedding_quantize='8,0' \
+                      base.preq_embedding_quantize=\'8,0\' \
                       model.use_sdpa_with_kv_cache=true \
                       model.use_kv_cache=true \
                       backend.xnnpack.enabled=true \
@@ -287,7 +287,7 @@ jobs:
                       backend.xnnpack.extended_ops=true \
                       quantization.qmode=8da4w \
                       quantization.group_size=32 \
-                      quantization.embedding_quantize='8,0' \
+                      quantization.embedding_quantize=\'8,0\' \
                       base.metadata='{"get_bos_id":128000,"get_eos_ids":[128009,128001]}' \
                       export.output_name="${OUT_ET_MODEL_NAME}.pte"
                     ls -lh "${OUT_ET_MODEL_NAME}.pte"
@@ -325,7 +325,7 @@ jobs:
                   backend.xnnpack.extended_ops=true \
                   quantization.qmode=8da4w \
                   quantization.group_size=32 \
-                  quantization.embedding_quantize='8,0' \
+                  quantization.embedding_quantize=\'8,0\' \
                   base.metadata='{"get_bos_id":151644,"get_eos_ids":[151645]}' \
                   export.output_name="${OUT_ET_MODEL_NAME}.pte"
                 ls -lh "${OUT_ET_MODEL_NAME}.pte"

.github/workflows/apple-perf.yml

@@ -237,7 +237,7 @@ jobs:
                 export.output_name="${OUT_ET_MODEL_NAME}.pte" \
                 model.use_kv_cache=true \
                 model.dtype_override=fp32 \
-                base.preq_embedding_quantize='8,0' \
+                base.preq_embedding_quantize=\'8,0\' \
                 quantization.use_spin_quant=native \
                 base.metadata='{"get_bos_id":128000,"get_eos_ids":[128009,128001]}'
               ls -lh "${OUT_ET_MODEL_NAME}.pte"
@@ -258,7 +258,7 @@ jobs:
                 base.use_lora=16 \
                 base.preq_mode="8da4w_output_8da8w" \
                 base.preq_group_size=32 \
-                base.preq_embedding_quantize='8,0' \
+                base.preq_embedding_quantize=\'8,0\' \
                 model.use_sdpa_with_kv_cache=true \
                 model.use_kv_cache=true \
                 backend.xnnpack.enabled=true \
@@ -296,7 +296,7 @@ jobs:
                 backend.xnnpack.extended_ops=true \
                 quantization.qmode=8da4w \
                 quantization.group_size=32 \
-                quantization.embedding_quantize='8,0' \
+                quantization.embedding_quantize=\'8,0\' \
                 base.metadata='{"get_bos_id":128000,"get_eos_ids":[128009,128001]}' \
                 export.output_name="${OUT_ET_MODEL_NAME}.pte"
               ls -lh "${OUT_ET_MODEL_NAME}.pte"
@@ -330,7 +330,7 @@ jobs:
                   backend.xnnpack.extended_ops=true \
                   quantization.qmode=8da4w \
                   quantization.group_size=32 \
-                  quantization.embedding_quantize='8,0' \
+                  quantization.embedding_quantize=\'8,0\' \
                   base.metadata='{"get_bos_id":151644,"get_eos_ids":[151645]}' \
                   export.output_name="${OUT_ET_MODEL_NAME}.pte"
                 ls -lh "${OUT_ET_MODEL_NAME}.pte"