这个repo是基于SimCSE的embedding model的训练。在SimCSE的基础上加入了用GLM last hidden states的作为encoder的初始化。
pip install transformers==4.23.1
pip install torch
pip install sentencepiece
pip install datasets
CNewsSum (https://dqwang122.github.io/projects/CNewSum/) 下载下来以后用preprocess_CNewSum.py去process,然后把生成的.txt 放进data/,就是现在的结果使用的training data。data/ 里面还有一个test_run_10k.txt 是一个sample data,按照这个格式,每一行是一句完整的话,然后train的时候会自动拿每一句话和dropout过后的这句话作为positive examples训练。
因为SimCSE使用了SentEval这个package去做evaluation,这里面的所有tasks都是英文的,所以需要自己下载一个中文的数据去做eval。这里我们简易的用一个中文的STS-B去做evaluation。数据已经放入SentEval里了,直接跑training的话会自动拿它做Dev set eval。
sh run_unsup_chinese_bert_cnews.sh
多卡的train一个没有GLM初始化的BERT encoder。model_name_or_path 那里选自己想用的Bert或者Roberta模型就行。如果是中文的话,可以用bert-base-chinese或者hfl/chinese-bert-wwm-ext。如果需要单卡训练的话就把distributed.launch那行comment了然后把上面那行comment out。
sh run_unsup_GLM_chinese_cnews.sh
这里和上面的区别在于最后一个argument --init_embeddings_model。这个argument 放你想要的huggingface上的GLM model的名字,比如THUDM/glm-large-chinese。有一个可能出现的问题是,在我的linux机器上,直接放这个名字从huggingface下载这个pretrained model缓存会有问题,在colab上好像不会出现这个问题。如果你也出现了可以先去git clone 你想要的model card到本地,然后在这里放model的本地位置从本地load就可以。
现在我还没在evaluation.py里加入带上GLM的inference,所以现在如果跑run_eval.sh只能用来evaluate没有GLM的model,或者是带GLM的model但是只用encoder去inference,这里还需要加一下。
Model | STS-B-Chinese Dev | STS-B-Chinese Test |
---|---|---|
bert-base-chinese-SimCSE | 0.74 | 0.68 |
bert-base-chinese-SimCSE-GLM-large | 0.52 | 0.42 |
拿pretrained bert-base-chinese 作为base-model, 在我们的数据上做contrastive learning
拿GLM-large-chinese的last hidden states,过一遍fully connected layer,作为一个bert-base-chinese encoder的输入,去做contrastive learning,然后在inference time的时候也用GLM做了初始化,再放入encoder得到的embedding。训练的时候GLM是freeze的。
**************************** Original SimCSE README ****************************
This repository contains the code and pre-trained models for our paper SimCSE: Simple Contrastive Learning of Sentence Embeddings.
**************************** Updates ****************************
- 8/31: Our paper has been accepted to EMNLP! Please check out our updated paper (with updated numbers and baselines).
- 5/12: We updated our unsupervised models with new hyperparameters and better performance.
- 5/10: We released our sentence embedding tool and demo code.
- 4/23: We released our training code.
- 4/20: We released our model checkpoints and evaluation code.
- 4/18: We released our paper. Check it out!
- Overview
- Getting Started
- Model List
- Use SimCSE with Huggingface
- Train SimCSE
- Bugs or Questions?
- Citation
- SimCSE Elsewhere
We propose a simple contrastive learning framework that works with both unlabeled and labeled data. Unsupervised SimCSE simply takes an input sentence and predicts itself in a contrastive learning framework, with only standard dropout used as noise. Our supervised SimCSE incorporates annotated pairs from NLI datasets into contrastive learning by using entailment
pairs as positives and contradiction
pairs as hard negatives. The following figure is an illustration of our models.
We provide an easy-to-use sentence embedding tool based on our SimCSE model (see our Wiki for detailed usage). To use the tool, first install the simcse
package from PyPI
pip install simcse
Or directly install it from our code
python setup.py install
Note that if you want to enable GPU encoding, you should install the correct version of PyTorch that supports CUDA. See PyTorch official website for instructions.
After installing the package, you can load our model by just two lines of code
from simcse import SimCSE
model = SimCSE("princeton-nlp/sup-simcse-bert-base-uncased")
See model list for a full list of available models.
Then you can use our model for encoding sentences into embeddings
embeddings = model.encode("A woman is reading.")
Compute the cosine similarities between two groups of sentences
sentences_a = ['A woman is reading.', 'A man is playing a guitar.']
sentences_b = ['He plays guitar.', 'A woman is making a photo.']
similarities = model.similarity(sentences_a, sentences_b)
Or build index for a group of sentences and search among them
sentences = ['A woman is reading.', 'A man is playing a guitar.']
model.build_index(sentences)
results = model.search("He plays guitar.")
We also support faiss, an efficient similarity search library. Just install the package following instructions here and simcse
will automatically use faiss
for efficient search.
WARNING: We have found that faiss
did not well support Nvidia AMPERE GPUs (3090 and A100). In that case, you should change to other GPUs or install the CPU version of faiss
package.
We also provide an easy-to-build demo website to show how SimCSE can be used in sentence retrieval. The code is based on DensePhrases' repo and demo (a lot of thanks to the authors of DensePhrases).
Our released models are listed as following. You can import these models by using the simcse
package or using HuggingFace's Transformers.
Note that the results are slightly better than what we have reported in the current version of the paper after adopting a new set of hyperparameters (for hyperparamters, see the training section).
Naming rules: unsup
and sup
represent "unsupervised" (trained on Wikipedia corpus) and "supervised" (trained on NLI datasets) respectively.
Besides using our provided sentence embedding tool, you can also easily import our models with HuggingFace's transformers
:
import torch
from scipy.spatial.distance import cosine
from transformers import AutoModel, AutoTokenizer
# Import our models. The package will take care of downloading the models automatically
tokenizer = AutoTokenizer.from_pretrained("princeton-nlp/sup-simcse-bert-base-uncased")
model = AutoModel.from_pretrained("princeton-nlp/sup-simcse-bert-base-uncased")
# Tokenize input texts
texts = [
"There's a kid on a skateboard.",
"A kid is skateboarding.",
"A kid is inside the house."
]
inputs = tokenizer(texts, padding=True, truncation=True, return_tensors="pt")
# Get the embeddings
with torch.no_grad():
embeddings = model(**inputs, output_hidden_states=True, return_dict=True).pooler_output
# Calculate cosine similarities
# Cosine similarities are in [-1, 1]. Higher means more similar
cosine_sim_0_1 = 1 - cosine(embeddings[0], embeddings[1])
cosine_sim_0_2 = 1 - cosine(embeddings[0], embeddings[2])
print("Cosine similarity between \"%s\" and \"%s\" is: %.3f" % (texts[0], texts[1], cosine_sim_0_1))
print("Cosine similarity between \"%s\" and \"%s\" is: %.3f" % (texts[0], texts[2], cosine_sim_0_2))
If you encounter any problem when directly loading the models by HuggingFace's API, you can also download the models manually from the above table and use model = AutoModel.from_pretrained({PATH TO THE DOWNLOAD MODEL})
.
In the following section, we describe how to train a SimCSE model by using our code.
First, install PyTorch by following the instructions from the official website. To faithfully reproduce our results, please use the correct 1.7.1
version corresponding to your platforms/CUDA versions. PyTorch version higher than 1.7.1
should also work. For example, if you use Linux and CUDA11 (how to check CUDA version), install PyTorch by the following command,
pip install torch==1.7.1+cu110 -f https://download.pytorch.org/whl/torch_stable.html
If you instead use CUDA <11
or CPU, install PyTorch by the following command,
pip install torch==1.7.1
Then run the following script to install the remaining dependencies,
pip install -r requirements.txt
Our evaluation code for sentence embeddings is based on a modified version of SentEval. It evaluates sentence embeddings on semantic textual similarity (STS) tasks and downstream transfer tasks. For STS tasks, our evaluation takes the "all" setting, and report Spearman's correlation. See our paper (Appendix B) for evaluation details.
Before evaluation, please download the evaluation datasets by running
cd SentEval/data/downstream/
bash download_dataset.sh
Then come back to the root directory, you can evaluate any transformers
-based pre-trained models using our evaluation code. For example,
python evaluation.py \
--model_name_or_path princeton-nlp/sup-simcse-bert-base-uncased \
--pooler cls \
--task_set sts \
--mode test
which is expected to output the results in a tabular format:
------ test ------
+-------+-------+-------+-------+-------+--------------+-----------------+-------+
| STS12 | STS13 | STS14 | STS15 | STS16 | STSBenchmark | SICKRelatedness | Avg. |
+-------+-------+-------+-------+-------+--------------+-----------------+-------+
| 75.30 | 84.67 | 80.19 | 85.40 | 80.82 | 84.26 | 80.39 | 81.58 |
+-------+-------+-------+-------+-------+--------------+-----------------+-------+
Arguments for the evaluation script are as follows,
--model_name_or_path
: The name or path of atransformers
-based pre-trained checkpoint. You can directly use the models in the above table, e.g.,princeton-nlp/sup-simcse-bert-base-uncased
.--pooler
: Pooling method. Now we supportcls
(default): Use the representation of[CLS]
token. A linear+activation layer is applied after the representation (it's in the standard BERT implementation). If you use supervised SimCSE, you should use this option.cls_before_pooler
: Use the representation of[CLS]
token without the extra linear+activation. If you use unsupervised SimCSE, you should take this option.avg
: Average embeddings of the last layer. If you use checkpoints of SBERT/SRoBERTa (paper), you should use this option.avg_top2
: Average embeddings of the last two layers.avg_first_last
: Average embeddings of the first and last layers. If you use vanilla BERT or RoBERTa, this works the best.
--mode
: Evaluation modetest
(default): The default test mode. To faithfully reproduce our results, you should use this option.dev
: Report the development set results. Note that in STS tasks, onlySTS-B
andSICK-R
have development sets, so we only report their numbers. It also takes a fast mode for transfer tasks, so the running time is much shorter than thetest
mode (though numbers are slightly lower).fasttest
: It is the same astest
, but with a fast mode so the running time is much shorter, but the reported numbers may be lower (only for transfer tasks).
--task_set
: What set of tasks to evaluate on (if set, it will override--tasks
)sts
(default): Evaluate on STS tasks, includingSTS 12~16
,STS-B
andSICK-R
. This is the most commonly-used set of tasks to evaluate the quality of sentence embeddings.transfer
: Evaluate on transfer tasks.full
: Evaluate on both STS and transfer tasks.na
: Manually set tasks by--tasks
.
--tasks
: Specify which dataset(s) to evaluate on. Will be overridden if--task_set
is notna
. See the code for a full list of tasks.
Data
For unsupervised SimCSE, we sample 1 million sentences from English Wikipedia; for supervised SimCSE, we use the SNLI and MNLI datasets. You can run data/download_wiki.sh
and data/download_nli.sh
to download the two datasets.
Training scripts
We provide example training scripts for both unsupervised and supervised SimCSE. In run_unsup_example.sh
, we provide a single-GPU (or CPU) example for the unsupervised version, and in run_sup_example.sh
we give a multiple-GPU example for the supervised version. Both scripts call train.py
for training. We explain the arguments in following:
--train_file
: Training file path. We support "txt" files (one line for one sentence) and "csv" files (2-column: pair data with no hard negative; 3-column: pair data with one corresponding hard negative instance). You can use our provided Wikipedia or NLI data, or you can use your own data with the same format.--model_name_or_path
: Pre-trained checkpoints to start with. For now we support BERT-based models (bert-base-uncased
,bert-large-uncased
, etc.) and RoBERTa-based models (RoBERTa-base
,RoBERTa-large
, etc.).--temp
: Temperature for the contrastive loss.--pooler_type
: Pooling method. It's the same as the--pooler_type
in the evaluation part.--mlp_only_train
: We have found that for unsupervised SimCSE, it works better to train the model with MLP layer but test the model without it. You should use this argument when training unsupervised SimCSE models.--hard_negative_weight
: If using hard negatives (i.e., there are 3 columns in the training file), this is the logarithm of the weight. For example, if the weight is 1, then this argument should be set as 0 (default value).--do_mlm
: Whether to use the MLM auxiliary objective. If True:--mlm_weight
: Weight for the MLM objective.--mlm_probability
: Masking rate for the MLM objective.
All the other arguments are standard Huggingface's transformers
training arguments. Some of the often-used arguments are: --output_dir
, --learning_rate
, --per_device_train_batch_size
. In our example scripts, we also set to evaluate the model on the STS-B development set (need to download the dataset following the evaluation section) and save the best checkpoint.
For results in the paper, we use Nvidia 3090 GPUs with CUDA 11. Using different types of devices or different versions of CUDA/other softwares may lead to slightly different performance.
Hyperparameters
We use the following hyperparamters for training SimCSE:
Unsup. BERT | Unsup. RoBERTa | Sup. | |
---|---|---|---|
Batch size | 64 | 512 | 512 |
Learning rate (base) | 3e-5 | 1e-5 | 5e-5 |
Learning rate (large) | 1e-5 | 3e-5 | 1e-5 |
Convert models
Our saved checkpoints are slightly different from Huggingface's pre-trained checkpoints. Run python simcse_to_huggingface.py --path {PATH_TO_CHECKPOINT_FOLDER}
to convert it. After that, you can evaluate it by our evaluation code or directly use it out of the box.
If you have any questions related to the code or the paper, feel free to email Tianyu ([email protected]
) and Xingcheng ([email protected]
). If you encounter any problems when using the code, or want to report a bug, you can open an issue. Please try to specify the problem with details so we can help you better and quicker!
Please cite our paper if you use SimCSE in your work:
@inproceedings{gao2021simcse,
title={{SimCSE}: Simple Contrastive Learning of Sentence Embeddings},
author={Gao, Tianyu and Yao, Xingcheng and Chen, Danqi},
booktitle={Empirical Methods in Natural Language Processing (EMNLP)},
year={2021}
}
We thank the community's efforts for extending SimCSE!
- Jianlin Su has provided a Chinese version of SimCSE.
- AK391 integrated to Huggingface Spaces with Gradio. See demo:
- Nils Reimers has implemented a
sentence-transformers
-based training code for SimCSE.