Initial Commit

README.md

@@ -1,3 +1,54 @@
----
-license: apache-2.0
----
+# ONNX version of `sentence-transformers/all-mpnet-base-v2`
+
+This is the OONX version of https://huggingface.co/sentence-transformers/all-mpnet-base-v2, examined that the produced embeddings are the same.
+
+Optmized for CPU usage.
+
+## Convert
+
+The same checkpoint can also be created by using the `convert.py` script.
+
+## Usage - `transformers`
+
+Exactly the same as in `sentence-transformers/all-mpnet-base-v2` except using `ORTModelForFeatureExtraction` from optimum.
+
+```
+pip install optimum[onnxruntime]
+```
+
+```{python}
+from transformers import AutoTokenizer
+from optimum.onnxruntime import ORTModelForFeatureExtraction
+import torch
+import torch.nn.functional as F
+
+# Mean Pooling - Take attention mask into account for correct averaging
+def mean_pooling(model_output, attention_mask):
+    token_embeddings = model_output[0] #First element of model_output contains all token embeddings
+    input_mask_expanded = attention_mask.unsqueeze(-1).expand(token_embeddings.size()).float()
+    return torch.sum(token_embeddings * input_mask_expanded, 1) / torch.clamp(input_mask_expanded.sum(1), min=1e-9)
+
+
+# Sentences we want sentence embeddings for
+sentences = ['This is an example sentence', 'Each sentence is converted']
+
+# Load model from HuggingFace Hub
+tokenizer = AutoTokenizer.from_pretrained('sentence-transformers/all-mpnet-base-v2')
+model = ORTModelForFeatureExtraction.from_pretrained('sentence-transformers/all-mpnet-base-v2')
+
+# Tokenize sentences
+encoded_input = tokenizer(sentences, padding=True, truncation=True, return_tensors='pt')
+
+# Compute token embeddings
+with torch.no_grad():
+    model_output = model(**encoded_input)
+
+# Perform pooling
+sentence_embeddings = mean_pooling(model_output, encoded_input['attention_mask'])
+
+# Normalize embeddings
+sentence_embeddings = F.normalize(sentence_embeddings, p=2, dim=1)
+
+print("Sentence embeddings:")
+print(sentence_embeddings)
+```

config.json

@@ -0,0 +1,23 @@
+{
+  "_name_or_path": "microsoft/mpnet-base",
+  "architectures": [
+    "MPNetForMaskedLM"
+  ],
+  "attention_probs_dropout_prob": 0.1,
+  "bos_token_id": 0,
+  "eos_token_id": 2,
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 768,
+  "initializer_range": 0.02,
+  "intermediate_size": 3072,
+  "layer_norm_eps": 1e-05,
+  "max_position_embeddings": 514,
+  "model_type": "mpnet",
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "pad_token_id": 1,
+  "relative_attention_num_buckets": 32,
+  "transformers_version": "4.8.2",
+  "vocab_size": 30527
+}

convert.py

@@ -0,0 +1,22 @@
+from transformers import AutoModel
+import torch
+
+max_seq_length = 384
+
+model = AutoModel.from_pretrained("sentence-transformers/all-mpnet-base-v2")
+model.eval()
+
+inputs = {
+    "input_ids": torch.ones(1, max_seq_length, dtype=torch.int64),
+    "attention_mask": torch.ones(1, max_seq_length, dtype=torch.int64),
+}
+
+symbolic_names = {0: 'batch_size', 1: 'max_seq_len'}
+
+torch.onnx.export(
+    model,args=tuple(inputs.values()),
+    f="model.onnx",
+    export_params=True, 
+    input_names=["input_ids", "attention_mask"], output_names=["last_hidden_state"],
+    dynamic_axes={"input_ids": symbolic_names, "attention_mask": symbolic_names}
+)

model.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:91c3b3d55e18d17ed4657d4cc9207940ae14b526caa9ad79e55ae90cdc6f08ec
+size 438158583

special_tokens_map.json

@@ -0,0 +1 @@
+{"bos_token": "<s>", "eos_token": "</s>", "unk_token": "[UNK]", "sep_token": "</s>", "pad_token": "<pad>", "cls_token": "<s>", "mask_token": {"content": "<mask>", "single_word": false, "lstrip": true, "rstrip": false, "normalized": false}}

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"}