upload files

1_Pooling/config.json

@@ -0,0 +1,7 @@
+{
+  "word_embedding_dimension": 512,
+  "pooling_mode_cls_token": false,
+  "pooling_mode_mean_tokens": true,
+  "pooling_mode_max_tokens": false,
+  "pooling_mode_mean_sqrt_len_tokens": false
+}

README.md

@@ -0,0 +1,137 @@
+---
+language:
+- tr
+pipeline_tag: sentence-similarity
+tags:
+- sentence-transformers
+- feature-extraction
+- sentence-similarity
+- transformers
+datasets:
+- nli_tr
+- emrecan/stsb-mt-turkish
+license: mit
+---
+
+# turkish-small-bert-uncased-mean-nli-stsb-tr
+
+This is a [sentence-transformers](https://www.SBERT.net) model: It maps sentences & paragraphs to a 512 dimensional dense vector space and can be used for tasks like clustering or semantic search.
+
+This model was adapted from [ytu-ce-cosmos/turkish-small-bert-uncased](https://huggingface.co/ytu-ce-cosmos/turkish-small-bert-uncased) and fine-tuned on these datasets:
+- [nli_tr](https://huggingface.co/datasets/nli_tr)
+- [emrecan/stsb-mt-turkish](https://huggingface.co/datasets/emrecan/stsb-mt-turkish)
+
+## Usage (Sentence-Transformers)
+
+Using this model becomes easy when you have [sentence-transformers](https://www.SBERT.net) installed:
+
+```
+pip install -U sentence-transformers
+```
+
+Then you can use the model like this:
+
+```python
+from sentence_transformers import SentenceTransformer
+sentences = ["Bu örnek bir cümle", "Her cümle dönüştürülür"]
+
+model = SentenceTransformer('atasoglu/turkish-small-bert-uncased-mean-nli-stsb-tr')
+embeddings = model.encode(sentences)
+print(embeddings)
+```
+
+
+
+## Usage (HuggingFace Transformers)
+Without [sentence-transformers](https://www.SBERT.net), you can use the model like this: First, you pass your input through the transformer model, then you have to apply the right pooling-operation on-top of the contextualized word embeddings.
+
+```python
+from transformers import AutoTokenizer, AutoModel
+import torch
+
+
+#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 = ["Bu örnek bir cümle", "Her cümle dönüştürülür"]
+
+# Load model from HuggingFace Hub
+tokenizer = AutoTokenizer.from_pretrained('atasoglu/turkish-small-bert-uncased-mean-nli-stsb-tr')
+model = AutoModel.from_pretrained('atasoglu/turkish-small-bert-uncased-mean-nli-stsb-tr')
+
+# 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. In this case, mean pooling.
+sentence_embeddings = mean_pooling(model_output, encoded_input['attention_mask'])
+
+print("Sentence embeddings:")
+print(sentence_embeddings)
+```
+
+
+
+## Evaluation Results
+
+Achieved results on the [STS-b](https://huggingface.co/datasets/emrecan/stsb-mt-turkish) test split are given below:
+
+```txt
+Cosine-Similarity :       Pearson: 0.7387 Spearman: 0.7244
+Manhattan-Distance:       Pearson: 0.7118 Spearman: 0.7156
+Euclidean-Distance:       Pearson: 0.7119 Spearman: 0.7155
+Dot-Product-Similarity:   Pearson: 0.7164 Spearman: 0.7081
+```
+
+## Training
+The model was trained with the parameters:
+
+**DataLoader**:
+
+`torch.utils.data.dataloader.DataLoader` of length 90 with parameters:
+```
+{'batch_size': 64, 'sampler': 'torch.utils.data.sampler.RandomSampler', 'batch_sampler': 'torch.utils.data.sampler.BatchSampler'}
+```
+
+**Loss**:
+
+`sentence_transformers.losses.CosineSimilarityLoss.CosineSimilarityLoss` 
+
+Parameters of the fit()-Method:
+```
+{
+    "epochs": 5,
+    "evaluation_steps": 45,
+    "evaluator": "sentence_transformers.evaluation.EmbeddingSimilarityEvaluator.EmbeddingSimilarityEvaluator",
+    "max_grad_norm": 1,
+    "optimizer_class": "<class 'torch.optim.adamw.AdamW'>",
+    "optimizer_params": {
+        "lr": 2e-05
+    },
+    "scheduler": "WarmupLinear",
+    "steps_per_epoch": null,
+    "warmup_steps": 45,
+    "weight_decay": 0.01
+}
+```
+
+
+## Full Model Architecture
+```
+SentenceTransformer(
+  (0): Transformer({'max_seq_length': 256, 'do_lower_case': False}) with Transformer model: BertModel 
+  (1): Pooling({'word_embedding_dimension': 512, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': True, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False})
+)
+```
+
+## Citing & Authors
+
+<!--- Describe where people can find more information -->

config.json

@@ -0,0 +1,25 @@
+{
+  "_name_or_path": "e5_b64_turkish_small_bert_uncased-mean-nli/",
+  "architectures": [
+    "BertModel"
+  ],
+  "attention_probs_dropout_prob": 0.1,
+  "classifier_dropout": null,
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 512,
+  "initializer_range": 0.02,
+  "intermediate_size": 2048,
+  "layer_norm_eps": 1e-12,
+  "max_position_embeddings": 512,
+  "model_type": "bert",
+  "num_attention_heads": 8,
+  "num_hidden_layers": 4,
+  "pad_token_id": 0,
+  "position_embedding_type": "absolute",
+  "torch_dtype": "float32",
+  "transformers_version": "4.28.0",
+  "type_vocab_size": 2,
+  "use_cache": true,
+  "vocab_size": 32000
+}

config_sentence_transformers.json

@@ -0,0 +1,7 @@
+{
+  "__version__": {
+    "sentence_transformers": "2.2.2",
+    "transformers": "4.28.0",
+    "pytorch": "2.1.0+cu121"
+  }
+}

modules.json

@@ -0,0 +1,14 @@
+[
+  {
+    "idx": 0,
+    "name": "0",
+    "path": "",
+    "type": "sentence_transformers.models.Transformer"
+  },
+  {
+    "idx": 1,
+    "name": "1",
+    "path": "1_Pooling",
+    "type": "sentence_transformers.models.Pooling"
+  }
+]

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f74195b8a9669f7049a5ebf54d48133741be60ce00ceb478e8a08728bb735530
+size 118109958

sentence_bert_config.json

@@ -0,0 +1,4 @@
+{
+  "max_seq_length": 256,
+  "do_lower_case": false
+}

special_tokens_map.json

@@ -0,0 +1,7 @@
+{
+  "cls_token": "[CLS]",
+  "mask_token": "[MASK]",
+  "pad_token": "[PAD]",
+  "sep_token": "[SEP]",
+  "unk_token": "[UNK]"
+}

tokenizer_config.json

@@ -0,0 +1,17 @@
+{
+  "clean_up_tokenization_spaces": true,
+  "cls_token": "[CLS]",
+  "do_basic_tokenize": true,
+  "do_lower_case": false,
+  "mask_token": "[MASK]",
+  "max_len": 512,
+  "model_max_length": 512,
+  "never_split": null,
+  "pad_token": "[PAD]",
+  "sep_token": "[SEP]",
+  "strip_accents": null,
+  "tokenize_chinese_chars": true,
+  "tokenizer_class": "BertTokenizer",
+  "truncation": true,
+  "unk_token": "[UNK]"
+}

training.py

@@ -0,0 +1,210 @@
+# # -*- coding: utf-8 -*-
+# """turkish-sentence-embedding.ipynb
+
+# Automatically generated by Colaboratory.
+
+# Original file is located at
+#     https://colab.research.google.com/drive/1jvsd0ZRXCjsd5-lH6EI7GaEYIjHN-6d8
+# """
+
+# import sys
+# import torch
+# if not torch.cuda.is_available():
+#     print("CUDA NOT FOUND!")
+#     sys.exit(0)
+
+from datasets import load_dataset
+# ds_multinli = load_dataset("nli_tr", "multinli_tr")
+# ds_snli = load_dataset("nli_tr", "snli_tr")
+ds_stsb = load_dataset("emrecan/stsb-mt-turkish")
+
+# """# ALLNLI Training"""
+
+# import math
+# from sentence_transformers import models, losses, datasets
+from sentence_transformers import LoggingHandler, SentenceTransformer, util, InputExample
+# from sentence_transformers.evaluation import EmbeddingSimilarityEvaluator
+import logging
+# from datetime import datetime
+# import sys
+# import os
+# import gzip
+# import csv
+# import random
+
+# #### Just some code to print debug information to stdout
+logging.basicConfig(
+    format="%(asctime)s - %(message)s", datefmt="%Y-%m-%d %H:%M:%S", level=logging.INFO, handlers=[LoggingHandler()]
+)
+# #### /print debug information to stdout
+
+# model_name = "ytu-ce-cosmos/turkish-small-bert-uncased"
+train_batch_size = 64  # The larger you select this, the better the results (usually). But it requires more GPU memory
+max_seq_length = 75
+num_epochs = 5
+
+# # Save path of the model
+model_save_path = "e5_b64_turkish_small_bert_uncased-mean-nli"
+
+# # Here we define our SentenceTransformer model
+# word_embedding_model = models.Transformer(model_name, max_seq_length=max_seq_length).cuda()
+# pooling_model = models.Pooling(word_embedding_model.get_word_embedding_dimension(), pooling_mode="mean")
+# model = SentenceTransformer(modules=[word_embedding_model, pooling_model])
+
+# def add_to_samples(sent1, sent2, label):
+#     if sent1 not in train_data:
+#         train_data[sent1] = {"contradiction": set(), "entailment": set(), "neutral": set()}
+#     train_data[sent1][label].add(sent2)
+
+# """
+# 0: neutral
+# 1: entailment
+# 2: contradiction
+# """
+# id_to_label = {0: "entailment", 1: "neutral", 2: "contradiction"}
+
+# train_data = {}
+
+# nan_count = 0
+# ds_allnli_train = [ds_multinli["train"], ds_snli["train"]]
+# for ds in ds_allnli_train:
+#     for row in ds:
+#         sent1 = row["premise"].strip()
+#         sent2 = row["hypothesis"].strip()
+#         label = row["label"]
+#         label = id_to_label.get(label)
+#         if label:
+#             add_to_samples(sent1, sent2, label)
+#             add_to_samples(sent2, sent1, label)  # Also add the opposite
+#         else:
+#             nan_count += 1
+
+# print("total Nan:", nan_count)
+
+
+# train_samples = []
+# for sent1, others in train_data.items():
+#     if len(others["entailment"]) > 0 and len(others["contradiction"]) > 0:
+#         train_samples.append(
+#             InputExample(
+#                 texts=[sent1, random.choice(list(others["entailment"])), random.choice(list(others["contradiction"]))]
+#             )
+#         )
+#         train_samples.append(
+#             InputExample(
+#                 texts=[random.choice(list(others["entailment"])), sent1, random.choice(list(others["contradiction"]))]
+#             )
+#         )
+
+# logging.info("Train samples: {}".format(len(train_samples)))
+
+# train_dataloader = datasets.NoDuplicatesDataLoader(train_samples, batch_size=train_batch_size)
+
+
+# # Our training loss
+# train_loss = losses.MultipleNegativesRankingLoss(model)
+
+# logging.info("Read STSbenchmark dev dataset")
+# dev_samples = []
+# for row in ds_stsb["validation"]:
+#     score = float(row["score"]) / 5.0  # Normalize score to range 0 ... 1
+#     dev_samples.append(InputExample(texts=[row["sentence1"], row["sentence2"]], label=score))
+
+# dev_evaluator = EmbeddingSimilarityEvaluator.from_input_examples(
+#     dev_samples, batch_size=train_batch_size, name="sts-dev"
+# )
+
+# test_samples = []
+# for row in ds_stsb["test"]:
+#     score = float(row["score"]) / 5.0  # Normalize score to range 0 ... 1
+#     test_samples.append(InputExample(texts=[row["sentence1"], row["sentence2"]], label=score))
+
+
+# test_evaluator = EmbeddingSimilarityEvaluator.from_input_examples(
+#     test_samples, batch_size=train_batch_size, name="sts-test"
+# )
+
+# # Configure the training
+# warmup_steps = math.ceil(len(train_dataloader) * num_epochs * 0.1)  # 10% of train data for warm-up
+# logging.info("Warmup-steps: {}".format(warmup_steps))
+
+# print(test_evaluator(model))
+
+# model.fit(
+#     train_objectives=[(train_dataloader, train_loss)],
+#     evaluator=dev_evaluator,
+#     epochs=num_epochs,
+#     evaluation_steps=int(len(train_dataloader) * 0.1),
+#     warmup_steps=warmup_steps,
+#     output_path=model_save_path,
+#     use_amp=False,  # Set to True, if your GPU supports FP16 operations
+# )
+
+# ft_model = SentenceTransformer(model_save_path)
+# print(test_evaluator(ft_model, output_path=model_save_path))
+
+from torch.utils.data import DataLoader
+import math
+from sentence_transformers import SentenceTransformer, LoggingHandler, losses, util, InputExample
+from sentence_transformers.evaluation import EmbeddingSimilarityEvaluator
+import logging
+from datetime import datetime
+import os
+import gzip
+import csv
+
+#### /print debug information to stdout
+
+
+# Read the dataset
+
+# Load a pre-trained sentence transformer model
+model = SentenceTransformer(model_save_path, device="cuda")
+
+model_save_path = "e5_b64_turkish_small_bert_uncased-mean-nli-stsb"
+
+# model_save_path = (
+#     "output/training_stsbenchmark_continue_training-" + model_name.replace("/", "-") + "-" + datetime.now().strftime("%Y-%m-%d_%H-%M-%S")
+# )
+# Convert the dataset to a DataLoader ready for training
+logging.info("Read STSbenchmark train dataset")
+
+def generate_samples(split):
+    samples = []
+    for row in ds_stsb[split]:
+        score = float(row["score"]) / 5.0  # Normalize score to range 0 ... 1
+        samples.append(InputExample(texts=[row["sentence1"], row["sentence2"]], label=score))
+    return samples
+
+train_samples = generate_samples("train")
+dev_samples = generate_samples("validation")
+test_samples = generate_samples("test")
+
+train_dataloader = DataLoader(train_samples, shuffle=True, batch_size=train_batch_size)
+train_loss = losses.CosineSimilarityLoss(model=model)
+
+
+# Development set: Measure correlation between cosine score and gold labels
+logging.info("Read STSbenchmark dev dataset")
+evaluator = EmbeddingSimilarityEvaluator.from_input_examples(dev_samples, name="sts-dev")
+
+
+# Configure the training. We skip evaluation in this example
+warmup_steps = math.ceil(len(train_dataloader) * num_epochs * 0.1)  # 10% of train data for warm-up
+logging.info("Warmup-steps: {}".format(warmup_steps))
+
+test_evaluator = EmbeddingSimilarityEvaluator.from_input_examples(test_samples, name="sts-test")
+print(test_evaluator(model))
+
+model.fit(
+    train_objectives=[(train_dataloader, train_loss)],
+    evaluator=evaluator,
+    epochs=num_epochs,
+    evaluation_steps=int(len(train_dataloader) * 0.5),
+    # evaluation_steps=1000,
+    warmup_steps=warmup_steps,
+    output_path=model_save_path,
+)
+
+ft_model = SentenceTransformer(model_save_path)
+print(test_evaluator(ft_model, output_path=model_save_path))