Merge branch 'main' into dev/vjatoth-qti/recipes-qnn-gpu

vjatoth-qti · web-flow · commit 860e1495d3e5 · 2025-10-17T11:58:24.000+05:30
diff --git a/Qwen-Qwen2.5-1.5B-Instruct/QNN/config.json b/Qwen-Qwen2.5-1.5B-Instruct/QNN/config.json
@@ -75,7 +75,7 @@
             "quant_preprocess": true,
             "op_types_to_exclude": [ "GatherBlockQuantized", "GroupQueryAttention", "MatMulNBits" ],
             "save_as_external_data": true,
-            "extra_option": { "CalibStridedMinMax": 4 }
+            "extra_options": { "CalibStridedMinMax": 4 }
         },
         "sp": { "type": "SplitModel" },
         "st": { "type": "StaticLLM", "batch_size": 1, "context_length": 64 },
diff --git a/Qwen-Qwen2.5-1.5B-Instruct/aitk/qwen2_5_qnn_config.json b/Qwen-Qwen2.5-1.5B-Instruct/aitk/qwen2_5_qnn_config.json
@@ -113,7 +113,7 @@
                 "MatMulNBits"
             ],
             "save_as_external_data": true,
-            "extra_option": {
+            "extra_options": {
                 "CalibStridedMinMax": 4
             }
         },
diff --git a/Qwen-Qwen2.5-7B-Instruct/QNN/config.json b/Qwen-Qwen2.5-7B-Instruct/QNN/config.json
@@ -75,7 +75,7 @@
             "quant_preprocess": true,
             "op_types_to_exclude": [ "GatherBlockQuantized", "GroupQueryAttention", "MatMulNBits" ],
             "save_as_external_data": true,
-            "extra_option": { "CalibStridedMinMax": 4 }
+            "extra_options": { "CalibStridedMinMax": 4 }
         },
         "sp": { "type": "SplitModel" },
         "st": { "type": "StaticLLM", "batch_size": 1, "context_length": 64 },
diff --git a/deepseek-ai-DeepSeek-R1-Distill-Qwen-1.5B/aitk/deepseek_qnn_config.json b/deepseek-ai-DeepSeek-R1-Distill-Qwen-1.5B/aitk/deepseek_qnn_config.json
@@ -113,7 +113,7 @@
                 "MatMulNBits"
             ],
             "save_as_external_data": true,
-            "extra_option": {
+            "extra_options": {
                 "CalibStridedMinMax": 4
             }
         },
diff --git a/meta-llama-Llama-3.1-8B-Instruct/QNN/config.json b/meta-llama-Llama-3.1-8B-Instruct/QNN/config.json
@@ -75,7 +75,7 @@
             "quant_preprocess": true,
             "op_types_to_exclude": [ "GatherBlockQuantized", "GroupQueryAttention", "MatMulNBits" ],
             "save_as_external_data": true,
-            "extra_option": { "CalibStridedMinMax": 4 }
+            "extra_options": { "CalibStridedMinMax": 4 }
         },
         "sp": { "type": "SplitModel" },
         "st": { "type": "StaticLLM", "batch_size": 1, "context_length": 64 },
diff --git a/meta-llama-Llama-3.1-8B-Instruct/aitk/llama3_1_qnn_config.json b/meta-llama-Llama-3.1-8B-Instruct/aitk/llama3_1_qnn_config.json
@@ -113,7 +113,7 @@
                 "MatMulNBits"
             ],
             "save_as_external_data": true,
-            "extra_option": {
+            "extra_options": {
                 "CalibStridedMinMax": 4
             }
         },
diff --git a/meta-llama-Llama-3.2-1B-Instruct/aitk/llama3_2_qnn_config.json b/meta-llama-Llama-3.2-1B-Instruct/aitk/llama3_2_qnn_config.json
@@ -113,7 +113,7 @@
                 "MatMulNBits"
             ],
             "save_as_external_data": true,
-            "extra_option": {
+            "extra_options": {
                 "CalibStridedMinMax": 4
             }
         },
diff --git a/microsoft-Phi-3-mini-128k-instruct/QNN/config.json b/microsoft-Phi-3-mini-128k-instruct/QNN/config.json
@@ -75,7 +75,7 @@
             "quant_preprocess": true,
             "op_types_to_exclude": [ "GatherBlockQuantized", "GroupQueryAttention", "MatMulNBits" ],
             "save_as_external_data": true,
-            "extra_option": { "CalibStridedMinMax": 4 }
+            "extra_options": { "CalibStridedMinMax": 4 }
         },
         "sp": { "type": "SplitModel" },
         "st": { "type": "StaticLLM", "batch_size": 1, "context_length": 64 },
diff --git a/microsoft-Phi-3-mini-4k-instruct/QNN/config.json b/microsoft-Phi-3-mini-4k-instruct/QNN/config.json
@@ -75,7 +75,7 @@
             "quant_preprocess": true,
             "op_types_to_exclude": [ "GatherBlockQuantized", "GroupQueryAttention", "MatMulNBits" ],
             "save_as_external_data": true,
-            "extra_option": { "CalibStridedMinMax": 4 }
+            "extra_options": { "CalibStridedMinMax": 4 }
         },
         "sp": { "type": "SplitModel" },
         "st": { "type": "StaticLLM", "batch_size": 1, "context_length": 64 },
diff --git a/microsoft-Phi-3.5-mini-instruct/QNN/config.json b/microsoft-Phi-3.5-mini-instruct/QNN/config.json
@@ -73,7 +73,7 @@
             "quant_preprocess": true,
             "op_types_to_exclude": [ "GatherBlockQuantized", "GroupQueryAttention", "MatMulNBits" ],
             "save_as_external_data": true,
-            "extra_option": { "CalibStridedMinMax": 4 }
+            "extra_options": { "CalibStridedMinMax": 4 }
         },
         "sp": { "type": "SplitModel" },
         "st": { "type": "StaticLLM", "batch_size": 1, "context_length": 64 },
diff --git a/microsoft-Phi-3.5-mini-instruct/QNN/config_fp16.json b/microsoft-Phi-3.5-mini-instruct/QNN/config_fp16.json
@@ -79,7 +79,7 @@
             "quant_preprocess": true,
             "op_types_to_exclude": [ "Cast", "GatherBlockQuantized", "GroupQueryAttention", "MatMulNBits" ],
             "save_as_external_data": true,
-            "extra_option": { "CalibStridedMinMax": 4 }
+            "extra_options": { "CalibStridedMinMax": 4 }
         },
         "sp": { "type": "SplitModel" },
         "st": { "type": "StaticLLM", "batch_size": 1, "context_length": 64 },
diff --git a/microsoft-Phi-3.5-mini-instruct/aitk/phi3_5_qnn_config.json b/microsoft-Phi-3.5-mini-instruct/aitk/phi3_5_qnn_config.json
@@ -113,7 +113,7 @@
                 "MatMulNBits"
             ],
             "save_as_external_data": true,
-            "extra_option": {
+            "extra_options": {
                 "CalibStridedMinMax": 4
             }
         },

Original file line number	Diff line number	Diff line change
`@@ -113,7 +113,7 @@`
`113`	`113`	`"MatMulNBits"`
`114`	`114`	`],`
`115`	`115`	`"save_as_external_data": true,`
`116`		`- "extra_option": {`
	`116`	`+ "extra_options": {`
`117`	`117`	`"CalibStridedMinMax": 4`
`118`	`118`	`}`
`119`	`119`	`},`