Create README.md

README.md

@@ -0,0 +1,343 @@
+---
+language: 
+- multilingual
+- af
+- sq
+- ar
+- an
+- hy
+- ast
+- az
+- ba
+- eu
+- bar
+- be
+- bn
+- inc
+- bs
+- br
+- bg
+- my
+- ca
+- ceb
+- ce
+- zh
+- cv
+- hr
+- cs
+- da
+- nl
+- en
+- et
+- fi
+- fr
+- gl
+- ka
+- de
+- el
+- gu
+- ht
+- he
+- hi
+- hu
+- is
+- io
+- id
+- ga
+- it
+- ja
+- jv
+- kn
+- kk
+- ky
+- ko
+- la
+- lv
+- lt
+- roa
+- nds
+- lm
+- mk
+- mg
+- ms
+- ml
+- mr
+- mn
+- min
+- ne
+- new
+- nb
+- nn
+- oc
+- fa
+- pms
+- pl
+- pt
+- pa
+- ro
+- ru
+- sco
+- sr
+- hr
+- scn
+- sk
+- sl
+- aze
+- es
+- su
+- sw
+- sv
+- tl
+- tg
+- th
+- ta
+- tt
+- te
+- tr
+- uk
+- ud
+- uz
+- vi
+- vo
+- war
+- cy
+- fry
+- pnb
+- yo
+tags:
+- onnx
+- awesome-align
+- word-alignment
+- bert
+pipeline_tag: feature-extraction
+license: apache-2.0
+datasets:
+- wikipedia
+---
+# Awesome-Align mBERT (ONNX FP32)
+
+This repository contains an ONNX export of **bert-base-multilingual-cased** specifically optimized for word alignment using the **awesome-align** methodology.
+
+### Model Details
+- **Base Model:** `bert-base-multilingual-cased`
+- **Truncation:** This model is truncated to the **first 8 layers**. According to the [awesome-align research](https://github.com/neulab/awesome-align), Layer 8 provides the optimal "sweet spot" for extracting cross-lingual word embeddings for alignment.
+- **Format:** ONNX (Full precision FP32)
+- **Size:** ~596 MB
+
+### Usage
+This model is intended to be used with `onnxruntime` to extract embeddings for source and target sentences.  It is truncated to Layer 8, the optimal layer for cross-lingual feature extraction. Alignments are then calculated using Cosine Similarity and Mutual Argmax (Intersection).
+
+```python
+import numpy as np
+import onnxruntime as ort
+from transformers import AutoTokenizer
+
+# 1. Load Model and Tokenizer
+# For INT8: use "cstr/awesome-align-onnx-int8"
+model_id = "cstr/awesome-align-onnx" 
+session = ort.InferenceSession("model.onnx", providers=['CPUExecutionProvider'])
+tokenizer = AutoTokenizer.from_pretrained(model_id)
+
+def get_word_embeddings(words):
+    # Tokenize with subword mapping
+    encoded = tokenizer(words, is_split_into_words=True, return_tensors="np")
+    
+    # Track which subwords belong to which original word index
+    word_map = []
+    for i, w in enumerate(words):
+        sub_tokens = tokenizer.tokenize(w) or [tokenizer.unk_token]
+        word_map.extend([i] * len(sub_tokens))
+
+    # Run inference
+    outputs = session.run(None, {
+        "input_ids": encoded["input_ids"],
+        "attention_mask": encoded["attention_mask"]
+    })
+    
+    # Slicing: [Batch 0, remove CLS/SEP, all hidden features]
+    embeddings = outputs[0][0, 1:-1, :]
+    return embeddings, word_map
+
+def align(src_words, tgt_words):
+    # Get embeddings and maps
+    src_embeds, src_map = get_word_embeddings(src_words)
+    tgt_embeds, tgt_map = get_word_embeddings(tgt_words)
+
+    # Compute Cosine Similarity
+    src_norm = src_embeds / np.linalg.norm(src_embeds, axis=-1, keepdims=True)
+    tgt_norm = tgt_embeds / np.linalg.norm(tgt_embeds, axis=-1, keepdims=True)
+    similarity = np.dot(src_norm, tgt_norm.T)
+
+    # Mutual Argmax (Intersection) Logic
+    best_tgt_for_src = np.argmax(similarity, axis=1)
+    best_src_for_tgt = np.argmax(similarity, axis=0)
+
+    alignment = set()
+    for i, j in enumerate(best_tgt_for_src):
+        if best_src_for_tgt[j] == i:
+            alignment.add((src_map[i], tgt_map[j]))
+    
+    return sorted(list(alignment))
+
+# Example Usage
+src = ["the", "cat", "sat"]
+tgt = ["le", "chat", "assis"]
+links = align(src, tgt)
+
+print(f"Alignment Links: {links}")
+# Output: [(0, 0), (1, 1), (2, 2)]
+```
+
+### Technical Notes
+
+* **Subword Handling**: This script automatically handles BERT subword tokenization by mapping sub-tokens (e.g., `['ass', '##is']`) back to their parent word index.
+* **Layer 8 Extraction**: The ONNX model is pre-truncated; the output of `session.run` is the 768-dimensional hidden state of the 8th layer.
+* **Precision**: The Mutual Argmax logic ensures high-precision 1-to-1 alignments. For higher recall, you can implement the "Grow-Diag-Final" heuristic on the `similarity` matrix.
+
+Original model card follows:
+
+# BERT multilingual base model (cased)
+
+Pretrained model on the top 104 languages with the largest Wikipedia using a masked language modeling (MLM) objective.
+It was introduced in [this paper](https://arxiv.org/abs/1810.04805) and first released in
+[this repository](https://github.com/google-research/bert). This model is case sensitive: it makes a difference
+between english and English.
+
+Disclaimer: The team releasing BERT did not write a model card for this model so this model card has been written by
+the Hugging Face team.
+
+## Model description
+
+BERT is a transformers model pretrained on a large corpus of multilingual data in a self-supervised fashion. This means
+it was pretrained on the raw texts only, with no humans labelling them in any way (which is why it can use lots of
+publicly available data) with an automatic process to generate inputs and labels from those texts. More precisely, it
+was pretrained with two objectives:
+
+- Masked language modeling (MLM): taking a sentence, the model randomly masks 15% of the words in the input then run
+  the entire masked sentence through the model and has to predict the masked words. This is different from traditional
+  recurrent neural networks (RNNs) that usually see the words one after the other, or from autoregressive models like
+  GPT which internally mask the future tokens. It allows the model to learn a bidirectional representation of the
+  sentence.
+- Next sentence prediction (NSP): the models concatenates two masked sentences as inputs during pretraining. Sometimes
+  they correspond to sentences that were next to each other in the original text, sometimes not. The model then has to
+  predict if the two sentences were following each other or not.
+
+This way, the model learns an inner representation of the languages in the training set that can then be used to
+extract features useful for downstream tasks: if you have a dataset of labeled sentences for instance, you can train a
+standard classifier using the features produced by the BERT model as inputs.
+
+## Intended uses & limitations
+
+You can use the raw model for either masked language modeling or next sentence prediction, but it's mostly intended to
+be fine-tuned on a downstream task. See the [model hub](https://huggingface.co/models?filter=bert) to look for
+fine-tuned versions on a task that interests you.
+
+Note that this model is primarily aimed at being fine-tuned on tasks that use the whole sentence (potentially masked)
+to make decisions, such as sequence classification, token classification or question answering. For tasks such as text
+generation you should look at model like GPT2.
+
+### How to use
+
+You can use this model directly with a pipeline for masked language modeling:
+
+```python
+>>> from transformers import pipeline
+>>> unmasker = pipeline('fill-mask', model='bert-base-multilingual-cased')
+>>> unmasker("Hello I'm a [MASK] model.")
+
+[{'sequence': "[CLS] Hello I'm a model model. [SEP]",
+  'score': 0.10182085633277893,
+  'token': 13192,
+  'token_str': 'model'},
+ {'sequence': "[CLS] Hello I'm a world model. [SEP]",
+  'score': 0.052126359194517136,
+  'token': 11356,
+  'token_str': 'world'},
+ {'sequence': "[CLS] Hello I'm a data model. [SEP]",
+  'score': 0.048930276185274124,
+  'token': 11165,
+  'token_str': 'data'},
+ {'sequence': "[CLS] Hello I'm a flight model. [SEP]",
+  'score': 0.02036019042134285,
+  'token': 23578,
+  'token_str': 'flight'},
+ {'sequence': "[CLS] Hello I'm a business model. [SEP]",
+  'score': 0.020079681649804115,
+  'token': 14155,
+  'token_str': 'business'}]
+```
+
+Here is how to use this model to get the features of a given text in PyTorch:
+
+```python
+from transformers import BertTokenizer, BertModel
+tokenizer = BertTokenizer.from_pretrained('bert-base-multilingual-cased')
+model = BertModel.from_pretrained("bert-base-multilingual-cased")
+text = "Replace me by any text you'd like."
+encoded_input = tokenizer(text, return_tensors='pt')
+output = model(**encoded_input)
+```
+
+and in TensorFlow:
+
+```python
+from transformers import BertTokenizer, TFBertModel
+tokenizer = BertTokenizer.from_pretrained('bert-base-multilingual-cased')
+model = TFBertModel.from_pretrained("bert-base-multilingual-cased")
+text = "Replace me by any text you'd like."
+encoded_input = tokenizer(text, return_tensors='tf')
+output = model(encoded_input)
+```
+
+## Training data
+
+The BERT model was pretrained on the 104 languages with the largest Wikipedias. You can find the complete list
+[here](https://github.com/google-research/bert/blob/master/multilingual.md#list-of-languages).
+
+## Training procedure
+
+### Preprocessing
+
+The texts are lowercased and tokenized using WordPiece and a shared vocabulary size of 110,000. The languages with a
+larger Wikipedia are under-sampled and the ones with lower resources are oversampled. For languages like Chinese,
+Japanese Kanji and Korean Hanja that don't have space, a CJK Unicode block is added around every character. 
+
+The inputs of the model are then of the form:
+
+```
+[CLS] Sentence A [SEP] Sentence B [SEP]
+```
+
+With probability 0.5, sentence A and sentence B correspond to two consecutive sentences in the original corpus and in
+the other cases, it's another random sentence in the corpus. Note that what is considered a sentence here is a
+consecutive span of text usually longer than a single sentence. The only constrain is that the result with the two
+"sentences" has a combined length of less than 512 tokens.
+
+The details of the masking procedure for each sentence are the following:
+- 15% of the tokens are masked.
+- In 80% of the cases, the masked tokens are replaced by `[MASK]`.
+- In 10% of the cases, the masked tokens are replaced by a random token (different) from the one they replace.
+- In the 10% remaining cases, the masked tokens are left as is.
+
+
+### BibTeX entry and citation info
+
+```bibtex
+@article{DBLP:journals/corr/abs-1810-04805,
+  author    = {Jacob Devlin and
+               Ming{-}Wei Chang and
+               Kenton Lee and
+               Kristina Toutanova},
+  title     = {{BERT:} Pre-training of Deep Bidirectional Transformers for Language
+               Understanding},
+  journal   = {CoRR},
+  volume    = {abs/1810.04805},
+  year      = {2018},
+  url       = {http://arxiv.org/abs/1810.04805},
+  archivePrefix = {arXiv},
+  eprint    = {1810.04805},
+  timestamp = {Tue, 30 Oct 2018 20:39:56 +0100},
+  biburl    = {https://dblp.org/rec/journals/corr/abs-1810-04805.bib},
+  bibsource = {dblp computer science bibliography, https://dblp.org}
+}
+```