Upload 3 files

all-mpnet-base-v2.py

@@ -0,0 +1,47 @@
+import torch
+from transformers import AutoTokenizer, AutoModel
+
+from executorch.backends.xnnpack.partition.xnnpack_partitioner import XnnpackPartitioner
+from executorch.exir import to_edge_transform_and_lower
+from torch.export import export, Dim
+from executorch.runtime import Runtime
+
+model_name = f"all-mpnet-base-v2.pte"
+tokenizer = AutoTokenizer.from_pretrained("sentence-transformers/all-mpnet-base-v2")
+model = AutoModel.from_pretrained("sentence-transformers/all-mpnet-base-v2")
+model.eval()
+
+sentences = ["This is an example sentence"]
+encoded_input = tokenizer(sentences, padding=True, truncation=True, return_tensors="pt")
+executorch_input = (encoded_input["input_ids"], encoded_input["attention_mask"])
+
+# Export the model
+executorch_program = to_edge_transform_and_lower(
+    export(
+        model,
+        executorch_input,
+        dynamic_shapes={
+            "input_ids": {
+                1: Dim("tokens", min=1, max=384),
+            },
+            "attention_mask": {
+                1: Dim("tokens", min=1, max=384),
+            },
+        },
+    ),
+    partitioner=[XnnpackPartitioner()],
+).to_executorch()
+
+# Save the model
+with open(model_name, "wb") as file:
+    file.write(executorch_program.buffer)
+
+# Check the model output
+eager_reference_output = model(**encoded_input)
+
+runtime = Runtime.get()
+program = runtime.load_program(model_name)
+method = program.load_method("forward")
+output = method.execute(executorch_input)
+
+print(torch.allclose(output[0], eager_reference_output[0], rtol=1e-3, atol=1e-5))

tokenizer_config.json

@@ -0,0 +1 @@
+{"do_lower_case": true, "bos_token": "<s>", "eos_token": "</s>", "sep_token": "</s>", "cls_token": "<s>", "unk_token": "[UNK]", "pad_token": "<pad>", "mask_token": "<mask>", "tokenize_chinese_chars": true, "strip_accents": null, "model_max_length": 512, "special_tokens_map_file": null, "name_or_path": "microsoft/mpnet-base", "tokenizer_class": "MPNetTokenizer"}