From 7e55a65999bc7af87670a3a0d9e294cac32b9b92 Mon Sep 17 00:00:00 2001 From: "github-actions[bot]" Date: Thu, 23 Jul 2026 20:09:40 +0800 Subject: [PATCH 1/2] recipe(bert): add dell-research-harvard/lt-un-data-fine-fine-es (feature-extraction, sentence-similarity) Curated CPU float recipes (fp32 + fp16) for the Spanish BETO sentence embedder lt-un-data-fine-fine-es (model_type=bert, BertModel, hidden 768, vocab 31002). Ships feature-extraction and sentence-similarity under cpu/cpu/ (quant:null; no CPU quantized variant per repo convention). The default build emits w8a16; these float recipes supply the CPU fp32/fp16 convention. L2 PyTorch-vs-ONNX parity: sentence-embedding cosine 1.000000, retrieval ranking identical. Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com> --- .../cpu/feature-extraction_fp16_config.json | 66 +++++++++++++++++++ .../cpu/feature-extraction_fp32_config.json | 66 +++++++++++++++++++ .../cpu/sentence-similarity_fp16_config.json | 66 +++++++++++++++++++ .../cpu/sentence-similarity_fp32_config.json | 66 +++++++++++++++++++ 4 files changed, 264 insertions(+) create mode 100644 examples/recipes/dell-research-harvard_lt-un-data-fine-fine-es/cpu/cpu/feature-extraction_fp16_config.json create mode 100644 examples/recipes/dell-research-harvard_lt-un-data-fine-fine-es/cpu/cpu/feature-extraction_fp32_config.json create mode 100644 examples/recipes/dell-research-harvard_lt-un-data-fine-fine-es/cpu/cpu/sentence-similarity_fp16_config.json create mode 100644 examples/recipes/dell-research-harvard_lt-un-data-fine-fine-es/cpu/cpu/sentence-similarity_fp32_config.json diff --git a/examples/recipes/dell-research-harvard_lt-un-data-fine-fine-es/cpu/cpu/feature-extraction_fp16_config.json b/examples/recipes/dell-research-harvard_lt-un-data-fine-fine-es/cpu/cpu/feature-extraction_fp16_config.json new file mode 100644 index 000000000..659160eba --- /dev/null +++ b/examples/recipes/dell-research-harvard_lt-un-data-fine-fine-es/cpu/cpu/feature-extraction_fp16_config.json @@ -0,0 +1,66 @@ +{ + "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, + 31002 + ] + }, + { + "name": "attention_mask", + "dtype": "int32", + "shape": [ + 1, + 512 + ], + "value_range": [ + 0, + 2 + ] + }, + { + "name": "token_type_ids", + "dtype": "int32", + "shape": [ + 1, + 512 + ], + "value_range": [ + 0, + 2 + ] + } + ], + "output_tensors": [ + { + "name": "last_hidden_state" + } + ] + }, + "optim": { + "clamp_constant_values": true + }, + "quant": null, + "compile": null, + "loader": { + "task": "feature-extraction", + "model_class": "AutoModel", + "model_type": "bert" + } +} diff --git a/examples/recipes/dell-research-harvard_lt-un-data-fine-fine-es/cpu/cpu/feature-extraction_fp32_config.json b/examples/recipes/dell-research-harvard_lt-un-data-fine-fine-es/cpu/cpu/feature-extraction_fp32_config.json new file mode 100644 index 000000000..659160eba --- /dev/null +++ b/examples/recipes/dell-research-harvard_lt-un-data-fine-fine-es/cpu/cpu/feature-extraction_fp32_config.json @@ -0,0 +1,66 @@ +{ + "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, + 31002 + ] + }, + { + "name": "attention_mask", + "dtype": "int32", + "shape": [ + 1, + 512 + ], + "value_range": [ + 0, + 2 + ] + }, + { + "name": "token_type_ids", + "dtype": "int32", + "shape": [ + 1, + 512 + ], + "value_range": [ + 0, + 2 + ] + } + ], + "output_tensors": [ + { + "name": "last_hidden_state" + } + ] + }, + "optim": { + "clamp_constant_values": true + }, + "quant": null, + "compile": null, + "loader": { + "task": "feature-extraction", + "model_class": "AutoModel", + "model_type": "bert" + } +} diff --git a/examples/recipes/dell-research-harvard_lt-un-data-fine-fine-es/cpu/cpu/sentence-similarity_fp16_config.json b/examples/recipes/dell-research-harvard_lt-un-data-fine-fine-es/cpu/cpu/sentence-similarity_fp16_config.json new file mode 100644 index 000000000..6c1a31273 --- /dev/null +++ b/examples/recipes/dell-research-harvard_lt-un-data-fine-fine-es/cpu/cpu/sentence-similarity_fp16_config.json @@ -0,0 +1,66 @@ +{ + "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, + 31002 + ] + }, + { + "name": "attention_mask", + "dtype": "int32", + "shape": [ + 1, + 512 + ], + "value_range": [ + 0, + 2 + ] + }, + { + "name": "token_type_ids", + "dtype": "int32", + "shape": [ + 1, + 512 + ], + "value_range": [ + 0, + 2 + ] + } + ], + "output_tensors": [ + { + "name": "last_hidden_state" + } + ] + }, + "optim": { + "clamp_constant_values": true + }, + "quant": null, + "compile": null, + "loader": { + "task": "sentence-similarity", + "model_class": "AutoModel", + "model_type": "bert" + } +} diff --git a/examples/recipes/dell-research-harvard_lt-un-data-fine-fine-es/cpu/cpu/sentence-similarity_fp32_config.json b/examples/recipes/dell-research-harvard_lt-un-data-fine-fine-es/cpu/cpu/sentence-similarity_fp32_config.json new file mode 100644 index 000000000..6c1a31273 --- /dev/null +++ b/examples/recipes/dell-research-harvard_lt-un-data-fine-fine-es/cpu/cpu/sentence-similarity_fp32_config.json @@ -0,0 +1,66 @@ +{ + "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, + 31002 + ] + }, + { + "name": "attention_mask", + "dtype": "int32", + "shape": [ + 1, + 512 + ], + "value_range": [ + 0, + 2 + ] + }, + { + "name": "token_type_ids", + "dtype": "int32", + "shape": [ + 1, + 512 + ], + "value_range": [ + 0, + 2 + ] + } + ], + "output_tensors": [ + { + "name": "last_hidden_state" + } + ] + }, + "optim": { + "clamp_constant_values": true + }, + "quant": null, + "compile": null, + "loader": { + "task": "sentence-similarity", + "model_class": "AutoModel", + "model_type": "bert" + } +} From 86821d476bcfcc487dc993783557e49b0a930dfb Mon Sep 17 00:00:00 2001 From: "Shiyi Zheng (from Dev Box)" Date: Thu, 13 Aug 2026 00:31:48 +0800 Subject: [PATCH 2/2] recipe(bert): refresh dell Spanish encoder configs --- .../cpu/cpu/feature-extraction_fp16_config.json | 5 ++++- .../cpu/cpu/feature-extraction_fp32_config.json | 5 ++++- .../cpu/cpu/sentence-similarity_fp16_config.json | 5 ++++- .../cpu/cpu/sentence-similarity_fp32_config.json | 5 ++++- 4 files changed, 16 insertions(+), 4 deletions(-) diff --git a/examples/recipes/dell-research-harvard_lt-un-data-fine-fine-es/cpu/cpu/feature-extraction_fp16_config.json b/examples/recipes/dell-research-harvard_lt-un-data-fine-fine-es/cpu/cpu/feature-extraction_fp16_config.json index 659160eba..515917997 100644 --- a/examples/recipes/dell-research-harvard_lt-un-data-fine-fine-es/cpu/cpu/feature-extraction_fp16_config.json +++ b/examples/recipes/dell-research-harvard_lt-un-data-fine-fine-es/cpu/cpu/feature-extraction_fp16_config.json @@ -51,7 +51,10 @@ { "name": "last_hidden_state" } - ] + ], + "compatibility": { + "transformers_attention": "eager" + } }, "optim": { "clamp_constant_values": true diff --git a/examples/recipes/dell-research-harvard_lt-un-data-fine-fine-es/cpu/cpu/feature-extraction_fp32_config.json b/examples/recipes/dell-research-harvard_lt-un-data-fine-fine-es/cpu/cpu/feature-extraction_fp32_config.json index 659160eba..515917997 100644 --- a/examples/recipes/dell-research-harvard_lt-un-data-fine-fine-es/cpu/cpu/feature-extraction_fp32_config.json +++ b/examples/recipes/dell-research-harvard_lt-un-data-fine-fine-es/cpu/cpu/feature-extraction_fp32_config.json @@ -51,7 +51,10 @@ { "name": "last_hidden_state" } - ] + ], + "compatibility": { + "transformers_attention": "eager" + } }, "optim": { "clamp_constant_values": true diff --git a/examples/recipes/dell-research-harvard_lt-un-data-fine-fine-es/cpu/cpu/sentence-similarity_fp16_config.json b/examples/recipes/dell-research-harvard_lt-un-data-fine-fine-es/cpu/cpu/sentence-similarity_fp16_config.json index 6c1a31273..77ccfd279 100644 --- a/examples/recipes/dell-research-harvard_lt-un-data-fine-fine-es/cpu/cpu/sentence-similarity_fp16_config.json +++ b/examples/recipes/dell-research-harvard_lt-un-data-fine-fine-es/cpu/cpu/sentence-similarity_fp16_config.json @@ -51,7 +51,10 @@ { "name": "last_hidden_state" } - ] + ], + "compatibility": { + "transformers_attention": "eager" + } }, "optim": { "clamp_constant_values": true diff --git a/examples/recipes/dell-research-harvard_lt-un-data-fine-fine-es/cpu/cpu/sentence-similarity_fp32_config.json b/examples/recipes/dell-research-harvard_lt-un-data-fine-fine-es/cpu/cpu/sentence-similarity_fp32_config.json index 6c1a31273..77ccfd279 100644 --- a/examples/recipes/dell-research-harvard_lt-un-data-fine-fine-es/cpu/cpu/sentence-similarity_fp32_config.json +++ b/examples/recipes/dell-research-harvard_lt-un-data-fine-fine-es/cpu/cpu/sentence-similarity_fp32_config.json @@ -51,7 +51,10 @@ { "name": "last_hidden_state" } - ] + ], + "compatibility": { + "transformers_attention": "eager" + } }, "optim": { "clamp_constant_values": true