From 2b2c13df2c4ab33abdf32680d3481babf11c0015 Mon Sep 17 00:00:00 2001
From: Shiyi Zheng <shzhen@microsoft.com>
Date: Wed, 15 Jul 2026 17:57:14 +0800
Subject: [PATCH] recipe(opus-mt-fr-en): add CPU fp32 and fp16 configs

---
 .../cpu/translation_fp16_decoder_config.json  | 76 +++++++++++++++++++
 .../cpu/translation_fp16_encoder_config.json  | 53 +++++++++++++
 .../cpu/translation_fp32_decoder_config.json  | 54 +++++++++++++
 .../cpu/translation_fp32_encoder_config.json  | 45 +++++++++++
 4 files changed, 228 insertions(+)
 create mode 100644 examples/recipes/Helsinki-NLP_opus-mt-fr-en/cpu/cpu/translation_fp16_decoder_config.json
 create mode 100644 examples/recipes/Helsinki-NLP_opus-mt-fr-en/cpu/cpu/translation_fp16_encoder_config.json
 create mode 100644 examples/recipes/Helsinki-NLP_opus-mt-fr-en/cpu/cpu/translation_fp32_decoder_config.json
 create mode 100644 examples/recipes/Helsinki-NLP_opus-mt-fr-en/cpu/cpu/translation_fp32_encoder_config.json

diff --git a/examples/recipes/Helsinki-NLP_opus-mt-fr-en/cpu/cpu/translation_fp16_decoder_config.json b/examples/recipes/Helsinki-NLP_opus-mt-fr-en/cpu/cpu/translation_fp16_decoder_config.json
new file mode 100644
index 000000000..7762e86fa
--- /dev/null
+++ b/examples/recipes/Helsinki-NLP_opus-mt-fr-en/cpu/cpu/translation_fp16_decoder_config.json
@@ -0,0 +1,76 @@
+{
+  "export": {
+    "opset_version": 17,
+    "batch_size": 1,
+    "export_params": true,
+    "do_constant_folding": true,
+    "verbose": false,
+    "dynamo": false,
+    "enable_hierarchy_tags": true,
+    "clean_onnx": false,
+    "hierarchy_tag_format": "full",
+    "input_tensors": [
+      {"name": "decoder_input_ids", "dtype": "int32", "shape": [1, 1], "value_range": [0, 59514]},
+      {"name": "encoder_hidden_states", "dtype": "float32", "shape": [1, 512, 512], "value_range": [0, 1]},
+      {"name": "attention_mask", "dtype": "int64", "shape": [1, 512]},
+      {"name": "decoder_attention_mask", "dtype": "int64", "shape": [1, 512]},
+      {"name": "cache_position", "dtype": "int64", "shape": [1]},
+      {"name": "past_0_key", "dtype": "float32", "shape": [1, 8, 512, 64], "value_range": [0, 1]},
+      {"name": "past_0_value", "dtype": "float32", "shape": [1, 8, 512, 64], "value_range": [0, 1]},
+      {"name": "past_1_key", "dtype": "float32", "shape": [1, 8, 512, 64], "value_range": [0, 1]},
+      {"name": "past_1_value", "dtype": "float32", "shape": [1, 8, 512, 64], "value_range": [0, 1]},
+      {"name": "past_2_key", "dtype": "float32", "shape": [1, 8, 512, 64], "value_range": [0, 1]},
+      {"name": "past_2_value", "dtype": "float32", "shape": [1, 8, 512, 64], "value_range": [0, 1]},
+      {"name": "past_3_key", "dtype": "float32", "shape": [1, 8, 512, 64], "value_range": [0, 1]},
+      {"name": "past_3_value", "dtype": "float32", "shape": [1, 8, 512, 64], "value_range": [0, 1]},
+      {"name": "past_4_key", "dtype": "float32", "shape": [1, 8, 512, 64], "value_range": [0, 1]},
+      {"name": "past_4_value", "dtype": "float32", "shape": [1, 8, 512, 64], "value_range": [0, 1]},
+      {"name": "past_5_key", "dtype": "float32", "shape": [1, 8, 512, 64], "value_range": [0, 1]},
+      {"name": "past_5_value", "dtype": "float32", "shape": [1, 8, 512, 64], "value_range": [0, 1]}
+    ],
+    "output_tensors": [
+      {"name": "logits"},
+      {"name": "present_0_key"}, {"name": "present_0_value"},
+      {"name": "present_1_key"}, {"name": "present_1_value"},
+      {"name": "present_2_key"}, {"name": "present_2_value"},
+      {"name": "present_3_key"}, {"name": "present_3_value"},
+      {"name": "present_4_key"}, {"name": "present_4_value"},
+      {"name": "present_5_key"}, {"name": "present_5_value"}
+    ]
+  },
+  "optim": {
+    "clamp_constant_values": true,
+    "gelu_fusion": true,
+    "matmul_add_fusion": true,
+    "remove_isnan_in_attention_mask": true
+  },
+  "quant": {
+    "mode": "fp16",
+    "samples": 10,
+    "calibration_method": "minmax",
+    "weight_type": "uint8",
+    "activation_type": "uint8",
+    "per_channel": false,
+    "symmetric": false,
+    "weight_symmetric": null,
+    "activation_symmetric": null,
+    "save_calibration": false,
+    "distribution": "uniform",
+    "seed": null,
+    "calibration_load_path": null,
+    "calibration_save_path": null,
+    "op_types_to_quantize": null,
+    "nodes_to_exclude": null,
+    "task": "text2text-generation",
+    "model_id": "Helsinki-NLP/opus-mt-fr-en",
+    "model_type": "marian",
+    "fp16_keep_io_types": true,
+    "fp16_op_block_list": null
+  },
+  "compile": null,
+  "loader": {
+    "task": "text2text-generation",
+    "model_class": "MarianDecoderWrapper",
+    "model_type": "marian"
+  }
+}
diff --git a/examples/recipes/Helsinki-NLP_opus-mt-fr-en/cpu/cpu/translation_fp16_encoder_config.json b/examples/recipes/Helsinki-NLP_opus-mt-fr-en/cpu/cpu/translation_fp16_encoder_config.json
new file mode 100644
index 000000000..3e64f6310
--- /dev/null
+++ b/examples/recipes/Helsinki-NLP_opus-mt-fr-en/cpu/cpu/translation_fp16_encoder_config.json
@@ -0,0 +1,53 @@
+{
+  "export": {
+    "opset_version": 17,
+    "batch_size": 1,
+    "export_params": true,
+    "do_constant_folding": true,
+    "verbose": false,
+    "dynamo": false,
+    "enable_hierarchy_tags": true,
+    "clean_onnx": false,
+    "hierarchy_tag_format": "full",
+    "input_tensors": [
+      {"name": "input_ids", "dtype": "int32", "shape": [1, 512], "value_range": [0, 59514]},
+      {"name": "attention_mask", "dtype": "int32", "shape": [1, 512], "value_range": [0, 2]}
+    ],
+    "output_tensors": [{"name": "encoder_hidden_states"}]
+  },
+  "optim": {
+    "clamp_constant_values": true,
+    "gelu_fusion": true,
+    "matmul_add_fusion": true,
+    "remove_isnan_in_attention_mask": true
+  },
+  "quant": {
+    "mode": "fp16",
+    "samples": 10,
+    "calibration_method": "minmax",
+    "weight_type": "uint8",
+    "activation_type": "uint8",
+    "per_channel": false,
+    "symmetric": false,
+    "weight_symmetric": null,
+    "activation_symmetric": null,
+    "save_calibration": false,
+    "distribution": "uniform",
+    "seed": null,
+    "calibration_load_path": null,
+    "calibration_save_path": null,
+    "op_types_to_quantize": null,
+    "nodes_to_exclude": null,
+    "task": "feature-extraction",
+    "model_id": "Helsinki-NLP/opus-mt-fr-en",
+    "model_type": "marian",
+    "fp16_keep_io_types": true,
+    "fp16_op_block_list": null
+  },
+  "compile": null,
+  "loader": {
+    "task": "feature-extraction",
+    "model_class": "MarianEncoderWrapper",
+    "model_type": "marian"
+  }
+}
diff --git a/examples/recipes/Helsinki-NLP_opus-mt-fr-en/cpu/cpu/translation_fp32_decoder_config.json b/examples/recipes/Helsinki-NLP_opus-mt-fr-en/cpu/cpu/translation_fp32_decoder_config.json
new file mode 100644
index 000000000..4ca782f78
--- /dev/null
+++ b/examples/recipes/Helsinki-NLP_opus-mt-fr-en/cpu/cpu/translation_fp32_decoder_config.json
@@ -0,0 +1,54 @@
+{
+  "export": {
+    "opset_version": 17,
+    "batch_size": 1,
+    "export_params": true,
+    "do_constant_folding": true,
+    "verbose": false,
+    "dynamo": false,
+    "enable_hierarchy_tags": true,
+    "clean_onnx": false,
+    "hierarchy_tag_format": "full",
+    "input_tensors": [
+      {"name": "decoder_input_ids", "dtype": "int32", "shape": [1, 1], "value_range": [0, 59514]},
+      {"name": "encoder_hidden_states", "dtype": "float32", "shape": [1, 512, 512], "value_range": [0, 1]},
+      {"name": "attention_mask", "dtype": "int64", "shape": [1, 512]},
+      {"name": "decoder_attention_mask", "dtype": "int64", "shape": [1, 512]},
+      {"name": "cache_position", "dtype": "int64", "shape": [1]},
+      {"name": "past_0_key", "dtype": "float32", "shape": [1, 8, 512, 64], "value_range": [0, 1]},
+      {"name": "past_0_value", "dtype": "float32", "shape": [1, 8, 512, 64], "value_range": [0, 1]},
+      {"name": "past_1_key", "dtype": "float32", "shape": [1, 8, 512, 64], "value_range": [0, 1]},
+      {"name": "past_1_value", "dtype": "float32", "shape": [1, 8, 512, 64], "value_range": [0, 1]},
+      {"name": "past_2_key", "dtype": "float32", "shape": [1, 8, 512, 64], "value_range": [0, 1]},
+      {"name": "past_2_value", "dtype": "float32", "shape": [1, 8, 512, 64], "value_range": [0, 1]},
+      {"name": "past_3_key", "dtype": "float32", "shape": [1, 8, 512, 64], "value_range": [0, 1]},
+      {"name": "past_3_value", "dtype": "float32", "shape": [1, 8, 512, 64], "value_range": [0, 1]},
+      {"name": "past_4_key", "dtype": "float32", "shape": [1, 8, 512, 64], "value_range": [0, 1]},
+      {"name": "past_4_value", "dtype": "float32", "shape": [1, 8, 512, 64], "value_range": [0, 1]},
+      {"name": "past_5_key", "dtype": "float32", "shape": [1, 8, 512, 64], "value_range": [0, 1]},
+      {"name": "past_5_value", "dtype": "float32", "shape": [1, 8, 512, 64], "value_range": [0, 1]}
+    ],
+    "output_tensors": [
+      {"name": "logits"},
+      {"name": "present_0_key"}, {"name": "present_0_value"},
+      {"name": "present_1_key"}, {"name": "present_1_value"},
+      {"name": "present_2_key"}, {"name": "present_2_value"},
+      {"name": "present_3_key"}, {"name": "present_3_value"},
+      {"name": "present_4_key"}, {"name": "present_4_value"},
+      {"name": "present_5_key"}, {"name": "present_5_value"}
+    ]
+  },
+  "optim": {
+    "clamp_constant_values": true,
+    "gelu_fusion": true,
+    "matmul_add_fusion": true,
+    "remove_isnan_in_attention_mask": true
+  },
+  "quant": null,
+  "compile": null,
+  "loader": {
+    "task": "text2text-generation",
+    "model_class": "MarianDecoderWrapper",
+    "model_type": "marian"
+  }
+}
diff --git a/examples/recipes/Helsinki-NLP_opus-mt-fr-en/cpu/cpu/translation_fp32_encoder_config.json b/examples/recipes/Helsinki-NLP_opus-mt-fr-en/cpu/cpu/translation_fp32_encoder_config.json
new file mode 100644
index 000000000..41be91aa9
--- /dev/null
+++ b/examples/recipes/Helsinki-NLP_opus-mt-fr-en/cpu/cpu/translation_fp32_encoder_config.json
@@ -0,0 +1,45 @@
+{
+  "export": {
+    "opset_version": 17,
+    "batch_size": 1,
+    "export_params": true,
+    "do_constant_folding": true,
+    "verbose": false,
+    "dynamo": false,
+    "enable_hierarchy_tags": true,
+    "clean_onnx": false,
+    "hierarchy_tag_format": "full",
+    "input_tensors": [
+      {
+        "name": "input_ids",
+        "dtype": "int32",
+        "shape": [1, 512],
+        "value_range": [0, 59514]
+      },
+      {
+        "name": "attention_mask",
+        "dtype": "int32",
+        "shape": [1, 512],
+        "value_range": [0, 2]
+      }
+    ],
+    "output_tensors": [
+      {
+        "name": "encoder_hidden_states"
+      }
+    ]
+  },
+  "optim": {
+    "clamp_constant_values": true,
+    "gelu_fusion": true,
+    "matmul_add_fusion": true,
+    "remove_isnan_in_attention_mask": true
+  },
+  "quant": null,
+  "compile": null,
+  "loader": {
+    "task": "feature-extraction",
+    "model_class": "MarianEncoderWrapper",
+    "model_type": "marian"
+  }
+}