upload

1_Pooling/config.json

@@ -0,0 +1,7 @@
+{
+  "word_embedding_dimension": 384,
+  "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 +1,178 @@
+---
+pipeline_tag: sentence-similarity
+tags:
+- sentence-transformers
+- feature-extraction
+- sentence-similarity
+---
+
+# multi-qa-MiniLM-L6-cos-v1
+This is a [sentence-transformers](https://www.SBERT.net) model: It maps sentences & paragraphs to a 384 dimensional dense vector space and was designed for **semantic search**. It has been trained on 215M (question, answer) pairs from diverse sources. For an introduction to semantic search, have a look at: [SBERT.net - Semantic Search](https://www.sbert.net/examples/applications/semantic-search/README.html)
+
+
+## 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, util
+
+query = "How many people live in London?"
+docs = ["Around 9 Million people live in London", "London is known for its financial district"]
+
+#Load the model
+model = SentenceTransformer('sentence-transformers/multi-qa-MiniLM-L6-cos-v1')
+
+#Encode query and documents
+query_emb = model.encode(query)
+doc_emb = model.encode(docs)
+
+#Compute dot score between query and all document embeddings
+scores = util.dot_score(query_emb, doc_emb)[0].cpu().tolist()
+
+#Combine docs & scores
+doc_score_pairs = list(zip(docs, scores))
+
+#Sort by decreasing score
+doc_score_pairs = sorted(doc_score_pairs, key=lambda x: x[1], reverse=True)
+
+#Output passages & scores
+for doc, score in doc_score_pairs:
+    print(score, doc)
+```
+
+
+## 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 correct pooling-operation on-top of the contextualized word embeddings.
+
+```python
+from transformers import AutoTokenizer, AutoModel
+import torch
+import torch.nn.functional as F
+
+#Mean Pooling - Take average of all tokens
+def mean_pooling(model_output, attention_mask):
+    token_embeddings = model_output.last_hidden_state #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)
+
+
+#Encode text
+def encode(texts):
+    # Tokenize sentences
+    encoded_input = tokenizer(texts, padding=True, truncation=True, return_tensors='pt')
+
+    # Compute token embeddings
+    with torch.no_grad():
+        model_output = model(**encoded_input, return_dict=True)
+
+    # Perform pooling
+    embeddings = mean_pooling(model_output, encoded_input['attention_mask'])
+
+    # Normalize embeddings
+    embeddings = F.normalize(embeddings, p=2, dim=1)
+	
+    return embeddings
+
+
+# Sentences we want sentence embeddings for
+query = "How many people live in London?"
+docs = ["Around 9 Million people live in London", "London is known for its financial district"]
+
+# Load model from HuggingFace Hub
+tokenizer = AutoTokenizer.from_pretrained("sentence-transformers/multi-qa-MiniLM-L6-cos-v1")
+model = AutoModel.from_pretrained("sentence-transformers/multi-qa-MiniLM-L6-cos-v1")
+
+#Encode query and docs
+query_emb = encode(query)
+doc_emb = encode(docs)
+
+#Compute dot score between query and all document embeddings
+scores = torch.mm(query_emb, doc_emb.transpose(0, 1))[0].cpu().tolist()
+
+#Combine docs & scores
+doc_score_pairs = list(zip(docs, scores))
+
+#Sort by decreasing score
+doc_score_pairs = sorted(doc_score_pairs, key=lambda x: x[1], reverse=True)
+
+#Output passages & scores
+for doc, score in doc_score_pairs:
+    print(score, doc)
+```
+
+## Technical Details
+
+In the following some technical details how this model must be used:
+
+| Setting | Value |
+| --- | :---: |
+| Produces normalized embeddings | Yes |
+| Pooling-Method | Mean pooling |
+| Suitable score functions | dot-product (`util.dot_score`), cosine-similarity (`util.cos_sim`), or euclidean distance |
+
+Note: When loaded with `sentence-transformers`, this model produces normalized embeddings with length 1. In that case, dot-product and cosine-similarity are equivalent. dot-product is preferred as it is faster. Euclidean distance is proportional to dot-product and can also be used.
+
+----
+
+
+## Background
+
+The project aims to train sentence embedding models on very large sentence level datasets using a self-supervised 
+contrastive learning objective. We use a contrastive learning objective: given a sentence from the pair, the model should predict which out of a set of randomly sampled other sentences, was actually paired with it in our dataset.
+
+We developped this model during the 
+[Community week using JAX/Flax for NLP & CV](https://discuss.huggingface.co/t/open-to-the-community-community-week-using-jax-flax-for-nlp-cv/7104), 
+organized by Hugging Face. We developped this model as part of the project:
+[Train the Best Sentence Embedding Model Ever with 1B Training Pairs](https://discuss.huggingface.co/t/train-the-best-sentence-embedding-model-ever-with-1b-training-pairs/7354). We benefited from efficient hardware infrastructure to run the project: 7 TPUs v3-8, as well as intervention from Googles Flax, JAX, and Cloud team member about efficient deep learning frameworks.
+
+## Intended uses
+
+Our model is intented to be used for semantic search: It encodes queries / questions and text paragraphs in a dense vector space. It finds relevant documents for the given passages.
+
+Note that there is a limit of 512 word pieces: Text longer than that will be truncated. Further note that the model was just trained on input text up to 250 word pieces. It might not work well for longer text. 
+
+
+
+## Training procedure
+
+The full training script is accessible in this current repository: `train_script.py`.
+
+### Pre-training 
+
+We use the pretrained [`nreimers/MiniLM-L6-H384-uncased`](https://huggingface.co/nreimers/MiniLM-L6-H384-uncased) model. Please refer to the model card for more detailed information about the pre-training procedure.
+
+#### Training
+
+We use the concatenation from multiple datasets to fine-tune our model. In total we have about 215M (question, answer) pairs.
+We sampled each dataset given a weighted probability which configuration is detailed in the `data_config.json` file.
+
+The model was trained with [MultipleNegativesRankingLoss](https://www.sbert.net/docs/package_reference/losses.html#multiplenegativesrankingloss) using Mean-pooling, cosine-similarity as similarity function, and a scale of 20.
+
+
+
+
+| Dataset                    | Number of training tuples  |
+|--------------------------------------------------------|:--------------------------:|
+| [WikiAnswers](https://github.com/afader/oqa#wikianswers-corpus) Duplicate question pairs from WikiAnswers |  77,427,422 |
+| [PAQ](https://github.com/facebookresearch/PAQ) Automatically generated (Question, Paragraph) pairs for each paragraph in Wikipedia | 64,371,441 |
+| [Stack Exchange](https://huggingface.co/datasets/flax-sentence-embeddings/stackexchange_xml) (Title, Body) pairs from all StackExchanges  | 25,316,456 |
+| [Stack Exchange](https://huggingface.co/datasets/flax-sentence-embeddings/stackexchange_xml) (Title, Answer) pairs from all StackExchanges  |  21,396,559 |
+| [MS MARCO](https://microsoft.github.io/msmarco/) Triplets (query, answer, hard_negative) for 500k queries from Bing search engine |  17,579,773 |
+| [GOOAQ: Open Question Answering with Diverse Answer Types](https://github.com/allenai/gooaq) (query, answer) pairs for 3M Google queries and Google featured snippet  | 3,012,496 |
+| [Amazon-QA](http://jmcauley.ucsd.edu/data/amazon/qa/) (Question, Answer) pairs from Amazon product pages | 2,448,839 
+| [Yahoo Answers](https://www.kaggle.com/soumikrakshit/yahoo-answers-dataset) (Title, Answer) pairs from Yahoo Answers | 1,198,260 |
+| [Yahoo Answers](https://www.kaggle.com/soumikrakshit/yahoo-answers-dataset) (Question, Answer) pairs from Yahoo Answers | 681,164 |
+| [Yahoo Answers](https://www.kaggle.com/soumikrakshit/yahoo-answers-dataset) (Title, Question) pairs from Yahoo Answers | 659,896 |
+| [SearchQA](https://huggingface.co/datasets/search_qa) (Question, Answer) pairs for 140k questions, each with Top5 Google snippets on that question | 582,261 |
+| [ELI5](https://huggingface.co/datasets/eli5) (Question, Answer) pairs from Reddit ELI5 (explainlikeimfive) | 325,475 |
+| [Stack Exchange](https://huggingface.co/datasets/flax-sentence-embeddings/stackexchange_xml) Duplicate questions pairs (titles) | 304,525 |
+| [Quora Question Triplets](https://quoradata.quora.com/First-Quora-Dataset-Release-Question-Pairs) (Question, Duplicate_Question, Hard_Negative) triplets for Quora Questions Pairs dataset | 103,663 |
+| [Natural Questions (NQ)](https://ai.google.com/research/NaturalQuestions) (Question, Paragraph) pairs for 100k real Google queries with relevant Wikipedia paragraph | 100,231 |
+| [SQuAD2.0](https://rajpurkar.github.io/SQuAD-explorer/) (Question, Paragraph) pairs from SQuAD2.0 dataset |  87,599 |
+| [TriviaQA](https://huggingface.co/datasets/trivia_qa) (Question, Evidence) pairs | 73,346 |
+| **Total** | **214,988,242** |

config.json

@@ -0,0 +1,24 @@
+{
+  "_name_or_path": "nreimers/MiniLM-L6-H384-uncased",
+  "architectures": [
+    "BertModel"
+  ],
+  "attention_probs_dropout_prob": 0.1,
+  "gradient_checkpointing": false,
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 384,
+  "initializer_range": 0.02,
+  "intermediate_size": 1536,
+  "layer_norm_eps": 1e-12,
+  "max_position_embeddings": 512,
+  "model_type": "bert",
+  "num_attention_heads": 12,
+  "num_hidden_layers": 6,
+  "pad_token_id": 0,
+  "position_embedding_type": "absolute",
+  "transformers_version": "4.8.2",
+  "type_vocab_size": 2,
+  "use_cache": true,
+  "vocab_size": 30522
+}

config_sentence_transformers.json

@@ -0,0 +1,7 @@
+{
+  "__version__": {
+    "sentence_transformers": "2.0.0",
+    "transformers": "4.6.1",
+    "pytorch": "1.8.1"
+  }
+}

data_config.json

@@ -0,0 +1,942 @@
+[
+    {
+        "name": "stackexchange_title_body/skeptics.stackexchange.com.jsonl.gz",
+        "lines": 10009,
+        "weight": 3
+    },
+    {
+        "name": "stackexchange_Title_Answer/islam.stackexchange.com.jsonl.gz",
+        "lines": 10052,
+        "weight": 3
+    },
+    {
+        "name": "stackexchange_Title_Answer/anime.stackexchange.com.jsonl.gz",
+        "lines": 10131,
+        "weight": 3
+    },
+    {
+        "name": "stackexchange_title_body/writers.stackexchange.com.jsonl.gz",
+        "lines": 10157,
+        "weight": 3
+    },
+    {
+        "name": "stackexchange_title_body/astronomy.stackexchange.com.jsonl.gz",
+        "lines": 10462,
+        "weight": 3
+    },
+    {
+        "name": "stackexchange_title_body/vi.stackexchange.com.jsonl.gz",
+        "lines": 10551,
+        "weight": 3
+    },
+    {
+        "name": "stackexchange_Title_Answer/french.stackexchange.com.jsonl.gz",
+        "lines": 10578,
+        "weight": 3
+    },
+    {
+        "name": "stackexchange_title_body/cstheory.stackexchange.com.jsonl.gz",
+        "lines": 10642,
+        "weight": 3
+    },
+    {
+        "name": "stackexchange_Title_Answer/civicrm.stackexchange.com.jsonl.gz",
+        "lines": 10648,
+        "weight": 3
+    },
+    {
+        "name": "stackexchange_Title_Answer/expressionengine.stackexchange.com.jsonl.gz",
+        "lines": 10742,
+        "weight": 3
+    },
+    {
+        "name": "stackexchange_title_body/engineering.stackexchange.com.jsonl.gz",
+        "lines": 10753,
+        "weight": 3
+    },
+    {
+        "name": "stackexchange_Title_Answer/history.stackexchange.com.jsonl.gz",
+        "lines": 10766,
+        "weight": 3
+    },
+    {
+        "name": "stackexchange_title_body/french.stackexchange.com.jsonl.gz",
+        "lines": 10794,
+        "weight": 3
+    },
+    {
+        "name": "stackexchange_Title_Answer/politics.stackexchange.com.jsonl.gz",
+        "lines": 11047,
+        "weight": 3
+    },
+    {
+        "name": "stackexchange_title_body/economics.stackexchange.com.jsonl.gz",
+        "lines": 11115,
+        "weight": 3
+    },
+    {
+        "name": "stackexchange_Title_Answer/craftcms.stackexchange.com.jsonl.gz",
+        "lines": 11236,
+        "weight": 3
+    },
+    {
+        "name": "stackexchange_title_body/anime.stackexchange.com.jsonl.gz",
+        "lines": 11444,
+        "weight": 3
+    },
+    {
+        "name": "stackexchange_Title_Answer/christianity.stackexchange.com.jsonl.gz",
+        "lines": 11498,
+        "weight": 3
+    },
+    {
+        "name": "stackexchange_Title_Answer/softwarerecs.stackexchange.com.jsonl.gz",
+        "lines": 11761,
+        "weight": 3
+    },
+    {
+        "name": "stackexchange_Title_Answer/boardgames.stackexchange.com.jsonl.gz",
+        "lines": 11805,
+        "weight": 3
+    },
+    {
+        "name": "stackexchange_title_body/islam.stackexchange.com.jsonl.gz",
+        "lines": 11853,
+        "weight": 3
+    },
+    {
+        "name": "stackexchange_title_body/expressionengine.stackexchange.com.jsonl.gz",
+        "lines": 11866,
+        "weight": 3
+    },
+    {
+        "name": "stackexchange_title_body/politics.stackexchange.com.jsonl.gz",
+        "lines": 11894,
+        "weight": 3
+    },
+    {
+        "name": "stackexchange_title_body/history.stackexchange.com.jsonl.gz",
+        "lines": 12021,
+        "weight": 3
+    },
+    {
+        "name": "stackexchange_title_body/christianity.stackexchange.com.jsonl.gz",
+        "lines": 12108,
+        "weight": 3
+    },
+    {
+        "name": "stackexchange_title_body/boardgames.stackexchange.com.jsonl.gz",
+        "lines": 12149,
+        "weight": 3
+    },
+    {
+        "name": "stackexchange_title_body/civicrm.stackexchange.com.jsonl.gz",
+        "lines": 12543,
+        "weight": 3
+    },
+    {
+        "name": "stackexchange_title_body/craftcms.stackexchange.com.jsonl.gz",
+        "lines": 12574,
+        "weight": 3
+    },
+    {
+        "name": "stackexchange_Title_Answer/networkengineering.stackexchange.com.jsonl.gz",
+        "lines": 12590,
+        "weight": 3
+    },
+    {
+        "name": "stackexchange_Title_Answer/space.stackexchange.com.jsonl.gz",
+        "lines": 12893,
+        "weight": 3
+    },
+    {
+        "name": "stackexchange_Title_Answer/quant.stackexchange.com.jsonl.gz",
+        "lines": 12933,
+        "weight": 3
+    },
+    {
+        "name": "stackexchange_Title_Answer/philosophy.stackexchange.com.jsonl.gz",
+        "lines": 13114,
+        "weight": 3
+    },
+    {
+        "name": "stackexchange_Title_Answer/gardening.stackexchange.com.jsonl.gz",
+        "lines": 13246,
+        "weight": 3
+    },
+    {
+        "name": "stackexchange_title_body/hinduism.stackexchange.com.jsonl.gz",
+        "lines": 13450,
+        "weight": 4
+    },
+    {
+        "name": "stackexchange_title_body/networkengineering.stackexchange.com.jsonl.gz",
+        "lines": 13454,
+        "weight": 4
+    },
+    {
+        "name": "stackexchange_Title_Answer/german.stackexchange.com.jsonl.gz",
+        "lines": 13733,
+        "weight": 4
+    },
+    {
+        "name": "stackexchange_title_body/german.stackexchange.com.jsonl.gz",
+        "lines": 13950,
+        "weight": 4
+    },
+    {
+        "name": "stackexchange_title_body/philosophy.stackexchange.com.jsonl.gz",
+        "lines": 14829,
+        "weight": 4
+    },
+    {
+        "name": "stackexchange_title_body/gardening.stackexchange.com.jsonl.gz",
+        "lines": 15136,
+        "weight": 4
+    },
+    {
+        "name": "stackexchange_title_body/space.stackexchange.com.jsonl.gz",
+        "lines": 15142,
+        "weight": 4
+    },
+    {
+        "name": "stackexchange_Title_Answer/bicycles.stackexchange.com.jsonl.gz",
+        "lines": 15708,
+        "weight": 4
+    },
+    {
+        "name": "stackexchange_Title_Answer/law.stackexchange.com.jsonl.gz",
+        "lines": 16133,
+        "weight": 4
+    },
+    {
+        "name": "stackexchange_Title_Answer/arduino.stackexchange.com.jsonl.gz",
+        "lines": 16281,
+        "weight": 4
+    },
+    {
+        "name": "stackexchange_title_body/bicycles.stackexchange.com.jsonl.gz",
+        "lines": 16353,
+        "weight": 4
+    },
+    {
+        "name": "stackexchange_Title_Answer/emacs.stackexchange.com.jsonl.gz",
+        "lines": 16830,
+        "weight": 4
+    },
+    {
+        "name": "stackexchange_title_body/quant.stackexchange.com.jsonl.gz",
+        "lines": 17261,
+        "weight": 4
+    },
+    {
+        "name": "stackexchange_Title_Answer/dsp.stackexchange.com.jsonl.gz",
+        "lines": 17430,
+        "weight": 4
+    },
+    {
+        "name": "stackexchange_Title_Answer/puzzling.stackexchange.com.jsonl.gz",
+        "lines": 17448,
+        "weight": 4
+    },
+    {
+        "name": "stackexchange_title_body/puzzling.stackexchange.com.jsonl.gz",
+        "lines": 17851,
+        "weight": 5
+    },
+    {
+        "name": "stackexchange_title_body/law.stackexchange.com.jsonl.gz",
+        "lines": 17941,
+        "weight": 5
+    },
+    {
+        "name": "stackexchange_Title_Answer/movies.stackexchange.com.jsonl.gz",
+        "lines": 18243,
+        "weight": 5
+    },
+    {
+        "name": "stackexchange_Title_Answer/mechanics.stackexchange.com.jsonl.gz",
+        "lines": 18613,
+        "weight": 5
+    },
+    {
+        "name": "stackexchange_Title_Answer/aviation.stackexchange.com.jsonl.gz",
+        "lines": 18755,
+        "weight": 5
+    },
+    {
+        "name": "stackexchange_Title_Answer/biology.stackexchange.com.jsonl.gz",
+        "lines": 19277,
+        "weight": 5
+    },
+    {
+        "name": "stackexchange_Title_Answer/crypto.stackexchange.com.jsonl.gz",
+        "lines": 19404,
+        "weight": 5
+    },
+    {
+        "name": "stackexchange_title_body/arduino.stackexchange.com.jsonl.gz",
+        "lines": 19553,
+        "weight": 5
+    },
+    {
+        "name": "stackexchange_Title_Answer/music.stackexchange.com.jsonl.gz",
+        "lines": 19936,
+        "weight": 5
+    },
+    {
+        "name": "stackexchange_title_body/aviation.stackexchange.com.jsonl.gz",
+        "lines": 20139,
+        "weight": 5
+    },
+    {
+        "name": "stackexchange_title_body/softwarerecs.stackexchange.com.jsonl.gz",
+        "lines": 20142,
+        "weight": 5
+    },
+    {
+        "name": "stackexchange_title_body/movies.stackexchange.com.jsonl.gz",
+        "lines": 20181,
+        "weight": 5
+    },
+    {
+        "name": "stackexchange_Title_Answer/datascience.stackexchange.com.jsonl.gz",
+        "lines": 20503,
+        "weight": 5
+    },
+    {
+        "name": "stackexchange_title_body/music.stackexchange.com.jsonl.gz",
+        "lines": 20636,
+        "weight": 5
+    },
+    {
+        "name": "stackexchange_Title_Answer/japanese.stackexchange.com.jsonl.gz",
+        "lines": 20948,
+        "weight": 5
+    },
+    {
+        "name": "stackexchange_title_body/emacs.stackexchange.com.jsonl.gz",
+        "lines": 21055,
+        "weight": 5
+    },
+    {
+        "name": "stackexchange_title_body/dsp.stackexchange.com.jsonl.gz",
+        "lines": 21252,
+        "weight": 5
+    },
+    {
+        "name": "stackexchange_title_body/japanese.stackexchange.com.jsonl.gz",
+        "lines": 22056,
+        "weight": 5
+    },
+    {
+        "name": "stackexchange_Title_Answer/bitcoin.stackexchange.com.jsonl.gz",
+        "lines": 22474,
+        "weight": 6
+    },
+    {
+        "name": "stackexchange_Title_Answer/cooking.stackexchange.com.jsonl.gz",
+        "lines": 22641,
+        "weight": 6
+    },
+    {
+        "name": "stackexchange_title_body/mechanics.stackexchange.com.jsonl.gz",
+        "lines": 22868,
+        "weight": 6
+    },
+    {
+        "name": "stackexchange_Title_Answer/photo.stackexchange.com.jsonl.gz",
+        "lines": 23204,
+        "weight": 6
+    },
+    {
+        "name": "stackexchange_title_body/crypto.stackexchange.com.jsonl.gz",
+        "lines": 23231,
+        "weight": 6
+    },
+    {
+        "name": "stackexchange_title_body/cooking.stackexchange.com.jsonl.gz",
+        "lines": 23705,
+        "weight": 6
+    },
+    {
+        "name": "stackexchange_title_body/photo.stackexchange.com.jsonl.gz",
+        "lines": 23753,
+        "weight": 6
+    },
+    {
+        "name": "stackexchange_Title_Answer/workplace.stackexchange.com.jsonl.gz",
+        "lines": 24012,
+        "weight": 6
+    },
+    {
+        "name": "stackexchange_Title_Answer/meta.stackoverflow.com.jsonl.gz",
+        "lines": 24044,
+        "weight": 6
+    },
+    {
+        "name": "stackexchange_Title_Answer/raspberrypi.stackexchange.com.jsonl.gz",
+        "lines": 24143,
+        "weight": 6
+    },
+    {
+        "name": "stackexchange_title_body/workplace.stackexchange.com.jsonl.gz",
+        "lines": 24189,
+        "weight": 6
+    },
+    {
+        "name": "stackexchange_title_body/biology.stackexchange.com.jsonl.gz",
+        "lines": 24447,
+        "weight": 6
+    },
+    {
+        "name": "stackexchange_Title_Answer/webapps.stackexchange.com.jsonl.gz",
+        "lines": 24867,
+        "weight": 6
+    },
+    {
+        "name": "stackexchange_title_body/bitcoin.stackexchange.com.jsonl.gz",
+        "lines": 25374,
+        "weight": 6
+    },
+    {
+        "name": "stackexchange_Title_Answer/judaism.stackexchange.com.jsonl.gz",
+        "lines": 26085,
+        "weight": 6
+    },
+    {
+        "name": "stackexchange_Title_Answer/ethereum.stackexchange.com.jsonl.gz",
+        "lines": 26124,
+        "weight": 6
+    },
+    {
+        "name": "stackexchange_Title_Answer/worldbuilding.stackexchange.com.jsonl.gz",
+        "lines": 26210,
+        "weight": 6
+    },
+    {
+        "name": "stackexchange_title_body/worldbuilding.stackexchange.com.jsonl.gz",
+        "lines": 26763,
+        "weight": 7
+    },
+    {
+        "name": "stackexchange_Title_Answer/chemistry.stackexchange.com.jsonl.gz",
+        "lines": 27061,
+        "weight": 7
+    },
+    {
+        "name": "stackexchange_title_body/datascience.stackexchange.com.jsonl.gz",
+        "lines": 27397,
+        "weight": 7
+    },
+    {
+        "name": "stackexchange_Title_Answer/graphicdesign.stackexchange.com.jsonl.gz",
+        "lines": 28083,
+        "weight": 7
+    },
+    {
+        "name": "stackexchange_Title_Answer/ux.stackexchange.com.jsonl.gz",
+        "lines": 28901,
+        "weight": 7
+    },
+    {
+        "name": "stackexchange_title_body/ux.stackexchange.com.jsonl.gz",
+        "lines": 29403,
+        "weight": 7
+    },
+    {
+        "name": "stackexchange_Title_Answer/money.stackexchange.com.jsonl.gz",
+        "lines": 29404,
+        "weight": 7
+    },
+    {
+        "name": "stackexchange_title_body/webapps.stackexchange.com.jsonl.gz",
+        "lines": 29697,
+        "weight": 7
+    },
+    {
+        "name": "stackexchange_Title_Answer/cs.stackexchange.com.jsonl.gz",
+        "lines": 30010,
+        "weight": 7
+    },
+    {
+        "name": "stackexchange_title_body/graphicdesign.stackexchange.com.jsonl.gz",
+        "lines": 30233,
+        "weight": 7
+    },
+    {
+        "name": "stackexchange_Title_Answer/webmasters.stackexchange.com.jsonl.gz",
+        "lines": 30370,
+        "weight": 7
+    },
+    {
+        "name": "stackexchange_title_body/raspberrypi.stackexchange.com.jsonl.gz",
+        "lines": 30625,
+        "weight": 7
+    },
+    {
+        "name": "stackexchange_title_body/money.stackexchange.com.jsonl.gz",
+        "lines": 32021,
+        "weight": 8
+    },
+    {
+        "name": "stackexchange_title_body/judaism.stackexchange.com.jsonl.gz",
+        "lines": 32028,
+        "weight": 8
+    },
+    {
+        "name": "stackexchange_Title_Answer/academia.stackexchange.com.jsonl.gz",
+        "lines": 32137,
+        "weight": 8
+    },
+    {
+        "name": "stackexchange_title_body/ethereum.stackexchange.com.jsonl.gz",
+        "lines": 32760,
+        "weight": 8
+    },
+    {
+        "name": "stackexchange_title_body/academia.stackexchange.com.jsonl.gz",
+        "lines": 34331,
+        "weight": 8
+    },
+    {
+        "name": "stackexchange_title_body/chemistry.stackexchange.com.jsonl.gz",
+        "lines": 34506,
+        "weight": 8
+    },
+    {
+        "name": "stackexchange_title_body/webmasters.stackexchange.com.jsonl.gz",
+        "lines": 34559,
+        "weight": 8
+    },
+    {
+        "name": "stackexchange_title_body/meta.stackoverflow.com.jsonl.gz",
+        "lines": 36456,
+        "weight": 9
+    },
+    {
+        "name": "stackexchange_Title_Answer/travel.stackexchange.com.jsonl.gz",
+        "lines": 36533,
+        "weight": 9
+    },
+    {
+        "name": "stackexchange_Title_Answer/android.stackexchange.com.jsonl.gz",
+        "lines": 38077,
+        "weight": 9
+    },
+    {
+        "name": "stackexchange_title_body/cs.stackexchange.com.jsonl.gz",
+        "lines": 38314,
+        "weight": 9
+    },
+    {
+        "name": "stackexchange_Title_Answer/gamedev.stackexchange.com.jsonl.gz",
+        "lines": 40154,
+        "weight": 10
+    },
+    {
+        "name": "stackexchange_Title_Answer/rpg.stackexchange.com.jsonl.gz",
+        "lines": 40435,
+        "weight": 10
+    },
+    {
+        "name": "stackexchange_title_body/travel.stackexchange.com.jsonl.gz",
+        "lines": 41227,
+        "weight": 10
+    },
+    {
+        "name": "stackexchange_Title_Answer/codereview.stackexchange.com.jsonl.gz",
+        "lines": 41748,
+        "weight": 10
+    },
+    {
+        "name": "stackexchange_title_body/rpg.stackexchange.com.jsonl.gz",
+        "lines": 42303,
+        "weight": 10
+    },
+    {
+        "name": "stackexchange_title_body/codereview.stackexchange.com.jsonl.gz",
+        "lines": 45765,
+        "weight": 11
+    },
+    {
+        "name": "stackexchange_title_body/gamedev.stackexchange.com.jsonl.gz",
+        "lines": 46485,
+        "weight": 11
+    },
+    {
+        "name": "stackexchange_Title_Answer/softwareengineering.stackexchange.com.jsonl.gz",
+        "lines": 51326,
+        "weight": 12
+    },
+    {
+        "name": "stackexchange_Title_Answer/security.stackexchange.com.jsonl.gz",
+        "lines": 51355,
+        "weight": 12
+    },
+    {
+        "name": "stackexchange_title_body/android.stackexchange.com.jsonl.gz",
+        "lines": 51608,
+        "weight": 12
+    },
+    {
+        "name": "stackexchange_Title_Answer/diy.stackexchange.com.jsonl.gz",
+        "lines": 52896,
+        "weight": 12
+    },
+    {
+        "name": "stackexchange_title_body/softwareengineering.stackexchange.com.jsonl.gz",
+        "lines": 53942,
+        "weight": 13
+    },
+    {
+        "name": "stackexchange_Title_Answer/blender.stackexchange.com.jsonl.gz",
+        "lines": 54153,
+        "weight": 13
+    },
+    {
+        "name": "stackexchange_Title_Answer/scifi.stackexchange.com.jsonl.gz",
+        "lines": 54805,
+        "weight": 13
+    },
+    {
+        "name": "stackexchange_title_body/security.stackexchange.com.jsonl.gz",
+        "lines": 58000,
+        "weight": 14
+    },
+    {
+        "name": "stackexchange_Title_Answer/mathematica.stackexchange.com.jsonl.gz",
+        "lines": 59895,
+        "weight": 14
+    },
+    {
+        "name": "stackexchange_title_body/diy.stackexchange.com.jsonl.gz",
+        "lines": 60083,
+        "weight": 14
+    },
+    {
+        "name": "stackexchange_Title_Answer/meta.stackexchange.com.jsonl.gz",
+        "lines": 60744,
+        "weight": 14
+    },
+    {
+        "name": "stackexchange_title_body/scifi.stackexchange.com.jsonl.gz",
+        "lines": 61528,
+        "weight": 14
+    },
+    {
+        "name": "stackexchange_Title_Answer/drupal.stackexchange.com.jsonl.gz",
+        "lines": 67817,
+        "weight": 16
+    },
+    {
+        "name": "stackexchange_Title_Answer/dba.stackexchange.com.jsonl.gz",
+        "lines": 71449,
+        "weight": 17
+    },
+    {
+        "name": "stackexchange_title_body/mathematica.stackexchange.com.jsonl.gz",
+        "lines": 73131,
+        "weight": 17
+    },
+    {
+        "name": "stackexchange_Title_Answer/ell.stackexchange.com.jsonl.gz",
+        "lines": 77892,
+        "weight": 18
+    },
+    {
+        "name": "stackexchange_Title_Answer/magento.stackexchange.com.jsonl.gz",
+        "lines": 79241,
+        "weight": 18
+    },
+    {
+        "name": "stackexchange_title_body/drupal.stackexchange.com.jsonl.gz",
+        "lines": 79717,
+        "weight": 18
+    },
+    {
+        "name": "stackexchange_Title_Answer/sharepoint.stackexchange.com.jsonl.gz",
+        "lines": 80420,
+        "weight": 19
+    },
+    {
+        "name": "stackexchange_title_body/blender.stackexchange.com.jsonl.gz",
+        "lines": 80766,
+        "weight": 19
+    },
+    {
+        "name": "stackexchange_title_body/dba.stackexchange.com.jsonl.gz",
+        "lines": 81871,
+        "weight": 19
+    },
+    {
+        "name": "stackexchange_Title_Answer/gaming.stackexchange.com.jsonl.gz",
+        "lines": 82887,
+        "weight": 19
+    },
+    {
+        "name": "stackexchange_title_body/ell.stackexchange.com.jsonl.gz",
+        "lines": 83271,
+        "weight": 19
+    },
+    {
+        "name": "stackexchange_title_body/meta.stackexchange.com.jsonl.gz",
+        "lines": 83510,
+        "weight": 19
+    },
+    {
+        "name": "stackexchange_Title_Answer/wordpress.stackexchange.com.jsonl.gz",
+        "lines": 83621,
+        "weight": 19
+    },
+    {
+        "name": "stackexchange_Title_Answer/mathoverflow.net.jsonl.gz",
+        "lines": 85289,
+        "weight": 20
+    },
+    {
+        "name": "stackexchange_Title_Answer/salesforce.stackexchange.com.jsonl.gz",
+        "lines": 87272,
+        "weight": 20
+    },
+    {
+        "name": "stackexchange_title_body/gaming.stackexchange.com.jsonl.gz",
+        "lines": 88912,
+        "weight": 21
+    },
+    {
+        "name": "stackexchange_Title_Answer/apple.stackexchange.com.jsonl.gz",
+        "lines": 92487,
+        "weight": 21
+    },
+    {
+        "name": "stackexchange_title_body/sharepoint.stackexchange.com.jsonl.gz",
+        "lines": 94011,
+        "weight": 22
+    },
+    {
+        "name": "stackexchange_title_body/magento.stackexchange.com.jsonl.gz",
+        "lines": 99991,
+        "weight": 23
+    },
+    {
+        "name": "stackexchange_Title_Answer/gis.stackexchange.com.jsonl.gz",
+        "lines": 100254,
+        "weight": 23
+    },
+    {
+        "name": "stackexchange_title_body/wordpress.stackexchange.com.jsonl.gz",
+        "lines": 100474,
+        "weight": 23
+    },
+    {
+        "name": "stackexchange_Title_Answer/english.stackexchange.com.jsonl.gz",
+        "lines": 100640,
+        "weight": 23
+    },
+    {
+        "name": "stackexchange_title_body/salesforce.stackexchange.com.jsonl.gz",
+        "lines": 105260,
+        "weight": 24
+    },
+    {
+        "name": "stackexchange_title_body/english.stackexchange.com.jsonl.gz",
+        "lines": 109522,
+        "weight": 25
+    },
+    {
+        "name": "stackexchange_title_body/apple.stackexchange.com.jsonl.gz",
+        "lines": 110622,
+        "weight": 25
+    },
+    {
+        "name": "stackexchange_Title_Answer/stats.stackexchange.com.jsonl.gz",
+        "lines": 115679,
+        "weight": 27
+    },
+    {
+        "name": "stackexchange_title_body/mathoverflow.net.jsonl.gz",
+        "lines": 120851,
+        "weight": 28
+    },
+    {
+        "name": "stackexchange_Title_Answer/electronics.stackexchange.com.jsonl.gz",
+        "lines": 129494,
+        "weight": 30
+    },
+    {
+        "name": "stackexchange_title_body/gis.stackexchange.com.jsonl.gz",
+        "lines": 131000,
+        "weight": 30
+    },
+    {
+        "name": "stackexchange_Title_Answer/physics.stackexchange.com.jsonl.gz",
+        "lines": 141230,
+        "weight": 32
+    },
+    {
+        "name": "stackexchange_title_body/electronics.stackexchange.com.jsonl.gz",
+        "lines": 143582,
+        "weight": 33
+    },
+    {
+        "name": "TriviaQA_pairs.jsonl.gz",
+        "lines": 73346,
+        "weight": 34
+    },
+    {
+        "name": "stackexchange_Title_Answer/unix.stackexchange.com.jsonl.gz",
+        "lines": 155414,
+        "weight": 36
+    },
+    {
+        "name": "stackexchange_Title_Answer/tex.stackexchange.com.jsonl.gz",
+        "lines": 171628,
+        "weight": 39
+    },
+    {
+        "name": "squad_pairs.jsonl.gz",
+        "lines": 87599,
+        "weight": 40
+    },
+    {
+        "name": "stackexchange_title_body/physics.stackexchange.com.jsonl.gz",
+        "lines": 173307,
+        "weight": 40
+    },
+    {
+        "name": "stackexchange_title_body/stats.stackexchange.com.jsonl.gz",
+        "lines": 173466,
+        "weight": 40
+    },
+    {
+        "name": "stackexchange_title_body/unix.stackexchange.com.jsonl.gz",
+        "lines": 185997,
+        "weight": 42
+    },
+    {
+        "name": "NQ-train_pairs.jsonl.gz",
+        "lines": 100231,
+        "weight": 46
+    },
+    {
+        "name": "stackexchange_title_body/tex.stackexchange.com.jsonl.gz",
+        "lines": 202954,
+        "weight": 46
+    },
+    {
+        "name": "quora_duplicates_triplets.jsonl.gz",
+        "lines": 103663,
+        "weight": 47
+    },
+    {
+        "name": "stackexchange_Title_Answer/serverfault.com.jsonl.gz",
+        "lines": 238507,
+        "weight": 54
+    },
+    {
+        "name": "stackexchange_Title_Answer/askubuntu.com.jsonl.gz",
+        "lines": 267135,
+        "weight": 61
+    },
+    {
+        "name": "stackexchange_title_body/serverfault.com.jsonl.gz",
+        "lines": 270904,
+        "weight": 62
+    },
+    {
+        "name": "stackexchange_duplicate_questions_title_title.jsonl.gz",
+        "lines": 304525,
+        "weight": 69
+    },
+    {
+        "name": "stackexchange_title_body/askubuntu.com.jsonl.gz",
+        "lines": 347925,
+        "weight": 79
+    },
+    {
+        "name": "stackexchange_Title_Answer/superuser.com.jsonl.gz",
+        "lines": 352610,
+        "weight": 80
+    },
+    {
+        "name": "stackexchange_title_body/superuser.com.jsonl.gz",
+        "lines": 435463,
+        "weight": 99
+    },
+    {
+        "name": "stackexchange_title_body/small_stackexchanges.jsonl.gz",
+        "lines": 448146,
+        "weight": 102
+    },
+    {
+        "name": "stackexchange_Title_Answer/small_stackexchanges.jsonl.gz",
+        "lines": 460256,
+        "weight": 104
+    },
+    {
+        "name": "eli5_question_answer.jsonl.gz",
+        "lines": 325475,
+        "weight": 147
+    },
+    {
+        "name": "yahoo_answers_title_question.jsonl.gz",
+        "lines": 659896,
+        "weight": 149
+    },
+    {
+        "name": "PAQ_pairs.jsonl.gz",
+        "lines": 64371441,
+        "weight": 150
+    },
+    {
+        "name": "WikiAnswers_pairs.jsonl.gz",
+        "lines": 77427422,
+        "weight": 150
+    },
+    {
+        "name": "stackexchange_Title_Answer/math.stackexchange.com.jsonl.gz",
+        "lines": 1100953,
+        "weight": 226
+    },
+    {
+        "name": "yahoo_answers_title_answer.jsonl.gz",
+        "lines": 1198260,
+        "weight": 226
+    },
+    {
+        "name": "stackexchange_title_body/math.stackexchange.com.jsonl.gz",
+        "lines": 1338443,
+        "weight": 226
+    },
+    {
+        "name": "stackexchange_Title_Answer/stackoverflow.com-Posts.jsonl.gz",
+        "lines": 15768211,
+        "weight": 226
+    },
+    {
+        "name": "stackexchange_title_body/stackoverflow.com-Posts.jsonl.gz",
+        "lines": 18562443,
+        "weight": 226
+    },
+    {
+        "name": "searchQA_question_top5_snippets_merged.jsonl.gz",
+        "lines": 582261,
+        "weight": 263
+    },
+    {
+        "name": "amazon-qa-train-pairs.jsonl.gz",
+        "lines": 2448839,
+        "weight": 451
+    },
+    {
+        "name": "gooaq_pairs.jsonl.gz",
+        "lines": 3012496,
+        "weight": 451
+    },
+    {
+        "name": "msmarco-query_passage_negative_v2.jsonl.gz",
+        "lines": 17579773,
+        "weight": 1000
+    }
+]

modules.json

@@ -0,0 +1,20 @@
+[
+  {
+    "idx": 0,
+    "name": "0",
+    "path": "",
+    "type": "sentence_transformers.models.Transformer"
+  },
+  {
+    "idx": 1,
+    "name": "1",
+    "path": "1_Pooling",
+    "type": "sentence_transformers.models.Pooling"
+  },
+  {
+    "idx": 2,
+    "name": "2",
+    "path": "2_Normalize",
+    "type": "sentence_transformers.models.Normalize"
+  }
+]

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:df507ec1743de52aa0a1b401183c5fd8ca18e846689421ea6c94cae014d9b26b
+size 90888945

sentence_bert_config.json

@@ -0,0 +1,4 @@
+{
+  "max_seq_length": 128,
+  "do_lower_case": false
+}

special_tokens_map.json

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

tokenizer_config.json

@@ -0,0 +1 @@
+{"do_lower_case": true, "unk_token": "[UNK]", "sep_token": "[SEP]", "pad_token": "[PAD]", "cls_token": "[CLS]", "mask_token": "[MASK]", "tokenize_chinese_chars": true, "strip_accents": null, "special_tokens_map_file": null, "name_or_path": "nreimers/MiniLM-L6-H384-uncased", "do_basic_tokenize": true, "never_split": null, "tokenizer_class": "BertTokenizer", "model_max_length": 512}

train_script.py

@@ -0,0 +1,361 @@
+"""
+Train script for a single file
+
+Need to set the TPU address first:
+export XRT_TPU_CONFIG="localservice;0;localhost:51011"
+"""
+
+import torch.multiprocessing as mp
+import threading
+import time
+import random
+import sys
+import argparse
+import gzip
+import json
+import logging
+import tqdm
+import torch
+from torch import nn
+from torch.utils.data import DataLoader
+import torch
+import torch_xla
+import torch_xla.core
+import torch_xla.core.functions
+import torch_xla.core.xla_model as xm
+import torch_xla.distributed.xla_multiprocessing as xmp
+import torch_xla.distributed.parallel_loader as pl
+import os
+from shutil import copyfile
+
+
+from transformers import (
+    AdamW,
+    AutoModel,
+    AutoTokenizer,
+    get_linear_schedule_with_warmup,
+    set_seed,
+)
+
+class AutoModelForSentenceEmbedding(nn.Module):
+    def __init__(self, model_name, tokenizer, args):
+        super(AutoModelForSentenceEmbedding, self).__init__()
+
+        assert args.pooling in ['mean', 'cls']
+
+        self.model = AutoModel.from_pretrained(model_name)
+        self.normalize = not args.no_normalize
+        self.tokenizer = tokenizer
+        self.pooling = args.pooling
+
+    def forward(self, **kwargs):
+        model_output = self.model(**kwargs)
+        if self.pooling == 'mean':
+            embeddings = self.mean_pooling(model_output, kwargs['attention_mask'])
+        elif self.pooling == 'cls':
+            embeddings = self.cls_pooling(model_output, kwargs['attention_mask'])
+
+        if self.normalize:
+            embeddings = torch.nn.functional.normalize(embeddings, p=2, dim=1)
+
+        return embeddings
+
+    def mean_pooling(self, 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)
+
+    def cls_pooling(self, model_output, attention_mask):
+        return model_output[0][:,0]
+
+    def save_pretrained(self, output_path):
+        if xm.is_master_ordinal():
+            self.tokenizer.save_pretrained(output_path)
+            self.model.config.save_pretrained(output_path)
+
+        xm.save(self.model.state_dict(), os.path.join(output_path, "pytorch_model.bin"))
+       
+
+
+
+def train_function(index, args, queue):
+    tokenizer = AutoTokenizer.from_pretrained(args.model)
+    model = AutoModelForSentenceEmbedding(args.model, tokenizer, args)
+    
+  
+    ### Train Loop
+    device = xm.xla_device()
+    model = model.to(device)
+
+    # Instantiate optimizer
+    optimizer = AdamW(params=model.parameters(), lr=2e-5, correct_bias=True)
+
+    lr_scheduler = get_linear_schedule_with_warmup(
+        optimizer=optimizer,
+        num_warmup_steps=500,
+        num_training_steps=args.steps,
+    )
+    
+    # Now we train the model
+    cross_entropy_loss = nn.CrossEntropyLoss()
+    max_grad_norm = 1
+
+    model.train()
+   
+    for global_step in tqdm.trange(args.steps, disable=not xm.is_master_ordinal()):
+        #### Get the batch data
+        batch = queue.get()
+        #print(index, "batch {}x{}".format(len(batch), ",".join([str(len(b)) for b in batch])))
+        
+
+        if len(batch[0]) == 2: #(anchor, positive)
+            text1 = tokenizer([b[0] for b in batch], return_tensors="pt", max_length=args.max_length_a, truncation=True, padding="max_length")
+            text2 = tokenizer([b[1] for b in batch], return_tensors="pt", max_length=args.max_length_b, truncation=True, padding="max_length")
+
+            ### Compute embeddings
+            embeddings_a = model(**text1.to(device))
+            embeddings_b = model(**text2.to(device))
+            
+            ### Gather all embedings 
+            embeddings_a = torch_xla.core.functions.all_gather(embeddings_a)
+            embeddings_b = torch_xla.core.functions.all_gather(embeddings_b)
+
+            ### Compute similarity scores 512 x 512
+            scores = torch.mm(embeddings_a, embeddings_b.transpose(0, 1)) * args.scale
+        
+            ### Compute cross-entropy loss
+            labels = torch.tensor(range(len(scores)), dtype=torch.long, device=embeddings_a.device)  # Example a[i] should match with b[i]
+            
+            ## Symmetric loss as in CLIP
+            loss = (cross_entropy_loss(scores, labels) + cross_entropy_loss(scores.transpose(0, 1), labels)) / 2
+
+        else:   #(anchor, positive, negative)
+            text1 = tokenizer([b[0] for b in batch], return_tensors="pt", max_length=args.max_length_a, truncation=True, padding="max_length")
+            text2 = tokenizer([b[1] for b in batch], return_tensors="pt", max_length=args.max_length_b, truncation=True, padding="max_length")
+            text3 = tokenizer([b[2] for b in batch], return_tensors="pt", max_length=args.max_length_b, truncation=True, padding="max_length")
+
+            embeddings_a  = model(**text1.to(device))
+            embeddings_b1 = model(**text2.to(device))
+            embeddings_b2 = model(**text3.to(device))
+
+            embeddings_a  = torch_xla.core.functions.all_gather(embeddings_a)
+            embeddings_b1 = torch_xla.core.functions.all_gather(embeddings_b1)
+            embeddings_b2 = torch_xla.core.functions.all_gather(embeddings_b2)
+
+            embeddings_b = torch.cat([embeddings_b1, embeddings_b2])
+
+            ### Compute similarity scores 512 x 1024
+            scores = torch.mm(embeddings_a, embeddings_b.transpose(0, 1)) * args.scale
+        
+            ### Compute cross-entropy loss
+            labels = torch.tensor(range(len(scores)), dtype=torch.long, device=embeddings_a.device)  # Example a[i] should match with b[i]
+            
+            ## One-way loss
+            loss = cross_entropy_loss(scores, labels)
+
+        
+        # Backward pass
+        optimizer.zero_grad()
+        loss.backward()
+        torch.nn.utils.clip_grad_norm_(model.parameters(), max_grad_norm)
+        
+        xm.optimizer_step(optimizer, barrier=True)
+        lr_scheduler.step()
+
+
+        #Save model
+        if (global_step+1) % args.save_steps == 0:
+            output_path = os.path.join(args.output, str(global_step+1))
+            xm.master_print("save model: "+output_path)
+            model.save_pretrained(output_path)
+          
+            
+    output_path = os.path.join(args.output, "final")
+    xm.master_print("save model final: "+ output_path)
+    model.save_pretrained(output_path)
+
+
+def produce_data(args, queue, filepaths, dataset_indices):
+    global_batch_size = args.batch_size*args.nprocs    #Global batch size
+    num_same_dataset = int(args.nprocs / args.datasets_per_batch)
+    print("producer", "global_batch_size", global_batch_size)
+    print("producer", "num_same_dataset", num_same_dataset)
+    
+    datasets = []
+    for filepath in filepaths:
+        if "reddit_" in filepath:       #Special dataset class for Reddit files
+            data_obj = RedditDataset(filepath)
+        else:
+            data_obj = Dataset(filepath)
+        datasets.append(iter(data_obj)) 
+    
+    # Store if dataset is in a 2 col or 3 col format
+    num_cols = {idx: len(next(dataset)) for idx, dataset in enumerate(datasets)}
+
+    while True:
+        texts_in_batch = set()
+        batch_format = None     #2 vs 3 col format for this batch
+        
+        #Add data from several sub datasets
+        for _ in range(args.datasets_per_batch):
+            valid_dataset = False   #Check that datasets have the same 2/3 col format
+            while not valid_dataset:
+                data_idx = random.choice(dataset_indices)
+                if batch_format is None:
+                    batch_format = num_cols[data_idx]
+                    valid_dataset = True
+                else:   #Check that this dataset has the same format
+                    valid_dataset = (batch_format == num_cols[data_idx])
+            
+            #Get data from this dataset
+            dataset = datasets[data_idx]
+            local_batch_size = args.batch_size
+            if batch_format == 3 and args.batch_size_triplets is not None:
+                local_batch_size = args.batch_size_triplets
+
+            for _ in range(num_same_dataset):
+                for _ in range(args.nprocs):
+                    batch_device = []   #A batch for one device
+                    while len(batch_device) < local_batch_size:
+                        sample = next(dataset)
+                        in_batch = False
+                        for text in sample:
+                            if text in texts_in_batch:
+                                in_batch = True
+                                break
+                        
+                        if not in_batch:
+                            for text in sample:
+                                texts_in_batch.add(text)
+                            batch_device.append(sample)
+
+                    queue.put(batch_device)
+                      
+
+class RedditDataset:
+    """
+    A class that handles the reddit data files
+    """
+    def __init__(self, filepath):
+        self.filepath = filepath
+
+    def __iter__(self):
+        while True:
+            with gzip.open(self.filepath, "rt") as fIn:
+                    for line in fIn:
+                        data = json.loads(line)
+
+                        if "response" in data and "context" in data:
+                            yield [data["response"], data["context"]]
+
+class Dataset:
+    """
+    A class that handles one dataset
+    """
+    def __init__(self, filepath):
+        self.filepath = filepath
+
+    def __iter__(self):
+        max_dataset_size = 20*1000*1000    #Cache small datasets in memory
+        dataset = []
+        data_format = None
+
+        while dataset is None or len(dataset) == 0:
+            with gzip.open(self.filepath, "rt") as fIn:
+                for line in fIn:
+                    data = json.loads(line)
+                    if isinstance(data, dict):
+                        data = data['texts']
+
+                    if data_format is None:
+                        data_format = len(data)
+                    
+                    #Ensure that all entries are of the same 2/3 col format
+                    assert len(data) == data_format
+
+                    if dataset is not None:
+                        dataset.append(data)
+                        if len(dataset) >= max_dataset_size:
+                            dataset = None
+
+                    yield data
+                
+        # Data loaded. Now stream to the queue
+        # Shuffle for each epoch
+        while True:
+            random.shuffle(dataset)
+            for data in dataset:
+                yield data
+                
+               
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    parser.add_argument('--model', default='nreimers/MiniLM-L6-H384-uncased')
+    parser.add_argument('--steps', type=int, default=2000)
+    parser.add_argument('--save_steps', type=int, default=10000)
+    parser.add_argument('--batch_size', type=int, default=64)
+    parser.add_argument('--batch_size_triplets', type=int, default=None)
+    parser.add_argument('--max_length_a', type=int, default=128)
+    parser.add_argument('--max_length_b', type=int, default=128)
+    parser.add_argument('--nprocs', type=int, default=8)
+    parser.add_argument('--datasets_per_batch', type=int, default=2, help="Number of datasets per batch")
+    parser.add_argument('--scale', type=float, default=20, help="Use 20 for cossim, and 1 when you work with unnormalized embeddings with dot product")
+    parser.add_argument('--no_normalize', action="store_true", default=False, help="If set: Embeddings are not normalized")
+    parser.add_argument('--pooling', default='mean')
+    parser.add_argument('--data_folder', default="/data", help="Folder with your dataset files")
+    parser.add_argument('data_config', help="A data_config.json file")
+    parser.add_argument('output')
+    args = parser.parse_args()
+
+    # Ensure num proc is devisible by datasets_per_batch
+    assert (args.nprocs % args.datasets_per_batch) == 0
+  
+
+    logging.info("Output: "+args.output)
+    if os.path.exists(args.output):
+        print("Output folder already exists.")
+        input("Continue?")
+
+    # Write train script to output path
+    os.makedirs(args.output, exist_ok=True)
+
+    data_config_path = os.path.join(args.output, 'data_config.json')
+    copyfile(args.data_config, data_config_path)
+
+    train_script_path = os.path.join(args.output, 'train_script.py')
+    copyfile(__file__, train_script_path)
+    with open(train_script_path, 'a') as fOut:
+        fOut.write("\n\n# Script was called via:\n#python " + " ".join(sys.argv))
+
+
+
+    #Load data config
+    with open(args.data_config) as fIn:
+        data_config = json.load(fIn)
+
+    queue = mp.Queue(maxsize=100*args.nprocs)
+    
+    filepaths = []
+    dataset_indices = []
+    for idx, data in enumerate(data_config):
+        filepaths.append(os.path.join(os.path.expanduser(args.data_folder), data['name']))
+        dataset_indices.extend([idx]*data['weight'])
+
+    # Start producer
+    p = mp.Process(target=produce_data, args=(args, queue, filepaths, dataset_indices))
+    p.start()
+
+    # Run training
+    print("Start processes:", args.nprocs)
+    xmp.spawn(train_function, args=(args, queue), nprocs=args.nprocs, start_method='fork')
+    print("Training done")
+    print("It might be that not all processes exit automatically. In that case you must manually kill this process.")
+    print("With 'pkill python' you can kill all remaining python processes")
+    p.kill()
+    exit()
+
+
+
+# Script was called via:
+#python train_many_data_files_v2.py --steps 200000 --batch_size 128 --model nreimers/MiniLM-L6-H384-uncased --max_length_a 64 --max_length_b 250 train_data_configs/multi-qa_v1.json output/multi-qa_v1-MiniLM-L6-mean_cos