initial

README.md

@@ -0,0 +1,37 @@
+Aggretriever is an encoder to aggregate both lexical and semantic text information into a single-vector dense vector for dense retrieval, which is finetued on MS MARCO corpus with BM25 negative sampling, following the approach described in [Aggretriever: A Simple Approach to Aggregate Textual Representation for Robust Dense Passage Retrieval](https://arxiv.org/abs/2208.00511).
+
+<p align="center">
+<img src="https://raw.githubusercontent.com/castorini/dhr/main/fig/aggretriever_teaser.png" width="600">
+</p>
+
+The associated GitHub repository for fine-tuning is available [here](https://github.com/castorini/dhr) and the reproduce from pyserini is [here]. The following variants are also available:
+
+Model | Initialization | MARCO Dev | Encoder Path
+|---|---|---|---|---|---
+aggretriever-distilbert | distilbert-base-uncased | 34.1 | [castorini/aggretriever-distilbert](https://huggingface.co/castorini/aggretriever-distilbert)
+aggretriever-cocondenser | Luyu/co-condenser-marco | 36.2 | [castorini/aggretriever-cocondenser](https://huggingface.co/castorini/aggretriever-cocondenser)
+
+## Usage (HuggingFace Transformers)
+Using the model directly available in HuggingFace transformers. We use the implemented Aggretriever from pyserini [here](https://github.com/castorini/pyserini/blob/master/pyserini/encode/_aggretriever.py).
+
+```python
+from pyserini.encode._aggretriever import AggretrieverQueryEncoder
+from pyserini.encode._aggretriever import AggretrieverDocumentEncoder
+
+model_name = '/store/scratch/s269lin/experiments/aggretriever/hf_model/aggretriever-cocondenser'
+query_encoder = AggretrieverQueryEncoder(model_name, device='cpu')
+context_encoder = AggretrieverDocumentEncoder(model_name, device='cpu')
+
+query =  ["Where was Marie Curie born?"]
+contexts = [
+    "Maria Sklodowska, later known as Marie Curie, was born on November 7, 1867.",
+    "Born in Paris on 15 May 1859, Pierre Curie was the son of Eugène Curie, a doctor of French Catholic origin from Alsace."
+]
+
+# Compute embeddings: take the last-layer hidden state of the [CLS] token
+query_emb = query_encoder.encode(query)
+ctx_emb = context_encoder.encode(contexts)
+# Compute similarity scores using dot product
+score1 = query_emb @ ctx_emb[0]  # 45.56658
+score2 = query_emb @ ctx_emb[1]  # 45.81762
+```

config.json

@@ -0,0 +1,33 @@
+{
+  "_name_or_path": "/store2/scratch/s269lin/Aggretriever/results/experiments/msmarco/coCondenser-Aggretriever",
+  "architectures": [
+    "AggretrieverEncoder"
+  ],
+  "attention_probs_dropout_prob": 0.1,
+  "classifier_dropout": null,
+  "gradient_checkpointing": false,
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 768,
+  "id2label": {
+    "0": "LABEL_0"
+  },
+  "initializer_range": 0.02,
+  "intermediate_size": 3072,
+  "label2id": {
+    "LABEL_0": 0
+  },
+  "layer_norm_eps": 1e-12,
+  "max_position_embeddings": 512,
+  "model_type": "bert",
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "output_hidden_states": true,
+  "pad_token_id": 0,
+  "position_embedding_type": "absolute",
+  "torch_dtype": "float32",
+  "transformers_version": "4.15.0",
+  "type_vocab_size": 2,
+  "use_cache": true,
+  "vocab_size": 30522
+}

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:99ed0d5d6a10d669ae95ccf41420dc6f91ef45b7954255c31db1324a5d40a745
+size 438540579

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, "do_basic_tokenize": true, "never_split": null, "unk_token": "[UNK]", "sep_token": "[SEP]", "pad_token": "[PAD]", "cls_token": "[CLS]", "mask_token": "[MASK]", "tokenize_chinese_chars": true, "strip_accents": null, "model_max_length": 512, "name_or_path": "Luyu/co-condenser-marco", "special_tokens_map_file": "/bos/tmp0/luyug/outputs/condenser/models/l2-s6-km-L128-e8-lr1e-4-b256/special_tokens_map.json", "tokenizer_file": null, "tokenizer_class": "BertTokenizer"}