Integrate with (Sentence) Transformers

README.md

@@ -1,11 +1,66 @@
 ---
 license: apache-2.0
 pipeline_tag: text-ranking
-library_name: lightning-ir
+language:
+- en
+library_name: sentence-transformers
 base_model:
 - google/electra-base-discriminator
+tags:
+- transformers
 ---
 
-This model was introduced in the paper [A Systematic Investigation of Distilling Large Language Models into Cross-Encoders for Passage Re-ranking](https://arxiv.org/abs/2405.07920).
+## Cross-Encoder for Text Ranking
 
-For code, examples and more, please visit https://github.com/webis-de/msmarco-llm-distillation.
+This model is a port of the [webis/monoelectra-base](https://huggingface.co/webis/monoelectra-base) model from [lightning-ir](https://github.com/webis-de/lightning-ir) to [Sentence Transformers](https://sbert.net/) and [Transformers](https://huggingface.co/docs/transformers). 
+
+The original model was introduced in the paper [A Systematic Investigation of Distilling Large Language Models into Cross-Encoders for Passage Re-ranking](https://arxiv.org/abs/2405.07920). See https://github.com/webis-de/rank-distillm for code used to train the original model.
+
+The model can be used as a reranker in a 2-stage "retrieve-rerank" pipeline, where it reorders passages returned by a retriever model (e.g. an embedding model or BM25) given some query. See [SBERT.net Retrieve & Re-rank](https://www.sbert.net/examples/applications/retrieve_rerank/README.html) for more details.
+
+## Usage with Sentence Transformers
+
+The usage is easy when you have [SentenceTransformers](https://www.sbert.net/) installed. 
+
+```bash
+pip install sentence-transformers
+```
+
+Then you can use the pre-trained model like this:
+
+```python
+from sentence_transformers import CrossEncoder
+
+model = CrossEncoder("cross-encoder/monoelectra-base", trust_remote_code=True)
+scores = model.predict([
+    ("How many people live in Berlin?", "Berlin had a population of 3,520,031 registered inhabitants in an area of 891.82 square kilometers."),
+    ("How many people live in Berlin?", "Berlin is well known for its museums."),
+])
+print(scores)
+# [ 8.607138 -4.320078]
+```
+
+## Usage with Transformers
+
+```python
+from transformers import AutoTokenizer, AutoModelForSequenceClassification
+import torch
+
+model = AutoModelForSequenceClassification.from_pretrained("cross-encoder/monoelectra-base", trust_remote_code=True)
+tokenizer = AutoTokenizer.from_pretrained("cross-encoder/monoelectra-base")
+
+features = tokenizer(
+    [
+        ["How many people live in Berlin?", "How many people live in Berlin?"],
+        ["Berlin has a population of 3,520,031 registered inhabitants in an area of 891.82 square kilometers.", "New York City is famous for the Metropolitan Museum of Art."],
+    ],
+    padding=True,
+    truncation=True,
+    return_tensors="pt",
+)
+
+model.eval()
+with torch.no_grad():
+    scores = model(**features).logits
+    print(scores)
+```

config.json

@@ -1,8 +1,11 @@
 {
   "architectures": [
-    "CrossEncoderElectraModel"
+    "WebisCrossEncoderForSequenceClassification"
   ],
   "attention_probs_dropout_prob": 0.1,
+  "auto_map": {
+    "AutoModelForSequenceClassification": "modeling.WebisCrossEncoderForSequenceClassification"
+  },
   "backbone_model_type": "electra",
   "classifier_dropout": null,
   "doc_length": 256,
@@ -10,23 +13,33 @@
   "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": "cross-encoder",
+  "model_type": "electra",
   "num_attention_heads": 12,
   "num_hidden_layers": 12,
   "pad_token_id": 0,
   "pooling_strategy": "first",
   "position_embedding_type": "absolute",
   "query_length": 32,
+  "sentence_transformers": {
+    "activation_fn": "torch.nn.modules.linear.Identity",
+    "version": "4.0.1"
+  },
   "summary_activation": "gelu",
   "summary_last_dropout": 0.1,
   "summary_type": "first",
   "summary_use_proj": true,
   "torch_dtype": "float32",
-  "transformers_version": "4.41.2",
+  "transformers_version": "4.49.0",
   "type_vocab_size": 2,
   "use_cache": true,
   "vocab_size": 30522

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:763a09add59ddfc02d87d3211d40374769b2c7060f92507a885f3917a33f7caa
-size 435592020
+oid sha256:5516a5f0510d6b44fc7415d3b283118f935c6438391e44e0850d079c0e644796
+size 435593564

modeling.py

@@ -0,0 +1,88 @@
+
+from typing import Optional, Tuple, Union
+import torch
+from torch.nn import BCEWithLogitsLoss, CrossEntropyLoss, MSELoss
+from transformers import ElectraPreTrainedModel, ElectraModel, ElectraConfig
+from transformers.modeling_outputs import SequenceClassifierOutput
+
+
+class WebisCrossEncoderForSequenceClassification(ElectraPreTrainedModel):
+    def __init__(self, config: ElectraConfig):
+        super().__init__(config)
+        self.num_labels = config.num_labels
+        self.config = config
+        self.electra = ElectraModel(config)
+        self.linear = torch.nn.Linear(config.hidden_size, config.num_labels, bias=False)
+
+        # Initialize weights and apply final processing
+        self.post_init()
+    
+    def forward(
+        self,
+        input_ids: Optional[torch.Tensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        token_type_ids: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.Tensor] = None,
+        head_mask: Optional[torch.Tensor] = None,
+        inputs_embeds: Optional[torch.Tensor] = None,
+        labels: Optional[torch.Tensor] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+    ) -> Union[Tuple[torch.Tensor], SequenceClassifierOutput]:
+        r"""
+        labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
+            Labels for computing the sequence classification/regression loss. Indices should be in `[0, ...,
+            config.num_labels - 1]`. If `config.num_labels == 1` a regression loss is computed (Mean-Square loss), If
+            `config.num_labels > 1` a classification loss is computed (Cross-Entropy).
+        """
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        discriminator_hidden_states = self.electra(
+            input_ids,
+            attention_mask=attention_mask,
+            token_type_ids=token_type_ids,
+            position_ids=position_ids,
+            head_mask=head_mask,
+            inputs_embeds=inputs_embeds,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+
+        sequence_output = discriminator_hidden_states[0]
+        logits = self.linear(sequence_output[:, 0, :])  # Take [CLS] token representation for classification
+
+        loss = None
+        if labels is not None:
+            if self.config.problem_type is None:
+                if self.num_labels == 1:
+                    self.config.problem_type = "regression"
+                elif self.num_labels > 1 and (labels.dtype == torch.long or labels.dtype == torch.int):
+                    self.config.problem_type = "single_label_classification"
+                else:
+                    self.config.problem_type = "multi_label_classification"
+
+            if self.config.problem_type == "regression":
+                loss_fct = MSELoss()
+                if self.num_labels == 1:
+                    loss = loss_fct(logits.squeeze(), labels.squeeze())
+                else:
+                    loss = loss_fct(logits, labels)
+            elif self.config.problem_type == "single_label_classification":
+                loss_fct = CrossEntropyLoss()
+                loss = loss_fct(logits.view(-1, self.num_labels), labels.view(-1))
+            elif self.config.problem_type == "multi_label_classification":
+                loss_fct = BCEWithLogitsLoss()
+                loss = loss_fct(logits, labels)
+
+        if not return_dict:
+            output = (logits,) + discriminator_hidden_states[1:]
+            return ((loss,) + output) if loss is not None else output
+
+        return SequenceClassifierOutput(
+            loss=loss,
+            logits=logits,
+            hidden_states=discriminator_hidden_states.hidden_states,
+            attentions=discriminator_hidden_states.attentions,
+        )