--- language: "en" tags: - knowledge-distillation datasets: - ms_marco --- # Margin-MSE Trained PreTTR We provide a retrieval trained DistilBert-based PreTTR model (https://arxiv.org/abs/2004.14255). Our model is trained with Margin-MSE using a 3 teacher BERT_Cat (concatenated BERT scoring) ensemble on MSMARCO-Passage. This instance can be used to **re-rank a candidate set**. The architecture is a 6-layer DistilBERT, split at layer 3, with an additional single linear layer at the end for scoring the CLS token. If you want to know more about our simple, yet effective knowledge distillation method for efficient information retrieval models for a variety of student architectures that is used for this model instance check out our paper: https://arxiv.org/abs/2010.02666 🎉 For more information, training data, source code, and a minimal usage example please visit: https://github.com/sebastian-hofstaetter/neural-ranking-kd ## Configuration - We split the DistilBERT in half at layer 3 ## Model Code ````python from transformers import * from transformers.models.distilbert.modeling_distilbert import * import math import torch from torch import nn as nn class PreTTRConfig(DistilBertConfig): join_layer_idx = 3 class PreTTR(DistilBertModel): ''' PreTTR changes the distilbert model from huggingface to be able to split query and document until a set layer, we skipped compression present in the original from: Efficient Document Re-Ranking for Transformers by Precomputing Term Representations MacAvaney, et al. https://arxiv.org/abs/2004.14255 ''' config_class = PreTTRConfig def __init__(self, config): super().__init__(config) self.transformer = SplitTransformer(config) # Encoder, we override the classes, but the names stay the same -> so it gets properly initialized self.embeddings = PosOffsetEmbeddings(config) # Embeddings self._classification_layer = torch.nn.Linear(self.config.hidden_size, 1, bias=False) self.join_layer_idx = config.join_layer_idx def forward( self, query, document, use_fp16: bool = False) -> torch.Tensor: with torch.cuda.amp.autocast(enabled=use_fp16): query_input_ids = query["input_ids"] query_attention_mask = query["attention_mask"] document_input_ids = document["input_ids"][:, 1:] document_attention_mask = document["attention_mask"][:, 1:] query_embs = self.embeddings(query_input_ids) # (bs, seq_length, dim) document_embs = self.embeddings(document_input_ids, query_input_ids.shape[-1]) # (bs, seq_length, dim) tfmr_output = self.transformer( query_embs=query_embs, query_mask=query_attention_mask, doc_embs=document_embs, doc_mask=document_attention_mask, join_layer_idx=self.join_layer_idx ) hidden_state = tfmr_output[0] score = self._classification_layer(hidden_state[:, 0, :]).squeeze() return score class PosOffsetEmbeddings(nn.Module): def __init__(self, config): super().__init__() self.word_embeddings = nn.Embedding(config.vocab_size, config.dim, padding_idx=config.pad_token_id) self.position_embeddings = nn.Embedding(config.max_position_embeddings, config.dim) if config.sinusoidal_pos_embds: create_sinusoidal_embeddings( n_pos=config.max_position_embeddings, dim=config.dim, out=self.position_embeddings.weight ) self.LayerNorm = nn.LayerNorm(config.dim, eps=1e-12) self.dropout = nn.Dropout(config.dropout) def forward(self, input_ids, pos_offset=0): """ Parameters ---------- input_ids: torch.tensor(bs, max_seq_length) The token ids to embed. Outputs ------- embeddings: torch.tensor(bs, max_seq_length, dim) The embedded tokens (plus position embeddings, no token_type embeddings) """ seq_length = input_ids.size(1) position_ids = torch.arange(seq_length, dtype=torch.long, device=input_ids.device) # (max_seq_length) position_ids = position_ids.unsqueeze(0).expand_as(input_ids) + pos_offset # (bs, max_seq_length) word_embeddings = self.word_embeddings(input_ids) # (bs, max_seq_length, dim) position_embeddings = self.position_embeddings(position_ids) # (bs, max_seq_length, dim) embeddings = word_embeddings + position_embeddings # (bs, max_seq_length, dim) embeddings = self.LayerNorm(embeddings) # (bs, max_seq_length, dim) embeddings = self.dropout(embeddings) # (bs, max_seq_length, dim) return embeddings class SplitTransformer(nn.Module): def __init__(self, config): super().__init__() self.n_layers = config.n_layers layer = TransformerBlock(config) self.layer = nn.ModuleList([copy.deepcopy(layer) for _ in range(config.n_layers)]) def forward(self, query_embs, query_mask, doc_embs, doc_mask, join_layer_idx, output_attentions=False, output_hidden_states=False): """ Parameters ---------- x: torch.tensor(bs, seq_length, dim) Input sequence embedded. attn_mask: torch.tensor(bs, seq_length) Attention mask on the sequence. Outputs ------- hidden_state: torch.tensor(bs, seq_length, dim) Sequence of hiddens states in the last (top) layer all_hidden_states: Tuple[torch.tensor(bs, seq_length, dim)] Tuple of length n_layers with the hidden states from each layer. Optional: only if output_hidden_states=True all_attentions: Tuple[torch.tensor(bs, n_heads, seq_length, seq_length)] Tuple of length n_layers with the attention weights from each layer Optional: only if output_attentions=True """ all_hidden_states = () all_attentions = () # # query / doc sep. # hidden_state_q = query_embs hidden_state_d = doc_embs for layer_module in self.layer[:join_layer_idx]: layer_outputs_q = layer_module( x=hidden_state_q, attn_mask=query_mask, head_mask=None, output_attentions=output_attentions ) hidden_state_q = layer_outputs_q[-1] layer_outputs_d = layer_module( x=hidden_state_d, attn_mask=doc_mask, head_mask=None, output_attentions=output_attentions ) hidden_state_d = layer_outputs_d[-1] # # combine # x = torch.cat([hidden_state_q, hidden_state_d], dim=1) attn_mask = torch.cat([query_mask, doc_mask], dim=1) # # combined # hidden_state = x for layer_module in self.layer[join_layer_idx:]: layer_outputs = layer_module( x=hidden_state, attn_mask=attn_mask, head_mask=None, output_attentions=output_attentions ) hidden_state = layer_outputs[-1] # Add last layer if output_hidden_states: all_hidden_states = all_hidden_states + (hidden_state,) outputs = (hidden_state,) if output_hidden_states: outputs = outputs + (all_hidden_states,) if output_attentions: outputs = outputs + (all_attentions,) return outputs # last-layer hidden state, (all hidden states), (all attentions) # # init the model & tokenizer (using the distilbert tokenizer) # tokenizer = AutoTokenizer.from_pretrained("distilbert-base-uncased") # honestly not sure if that is the best way to go, but it works :) model = PreTTR.from_pretrained("sebastian-hofstaetter/prettr-distilbert-split_at_3-margin_mse-T2-msmarco") ```` ## Effectiveness on MSMARCO Passage We trained our model on the MSMARCO standard ("small"-400K query) training triples with knowledge distillation with a batch size of 32 on a single consumer-grade GPU (11GB memory). For re-ranking we used the top-1000 BM25 results. ### MSMARCO-DEV Here, we use the larger 49K query DEV set (same range as the smaller 7K DEV set, minimal changes possible) | | MRR@10 | NDCG@10 | |----------------------------------|--------|---------| | BM25 | .194 | .241 | | **Margin-MSE PreTTR** (Re-ranking) | .386 | .447 | For more metrics, baselines, info and analysis, please see the paper: https://arxiv.org/abs/2010.02666 ## Limitations & Bias - The model inherits social biases from both DistilBERT and MSMARCO. - The model is only trained on relatively short passages of MSMARCO (avg. 60 words length), so it might struggle with longer text. ## Citation If you use our model checkpoint please cite our work as: ``` @misc{hofstaetter2020_crossarchitecture_kd, title={Improving Efficient Neural Ranking Models with Cross-Architecture Knowledge Distillation}, author={Sebastian Hofst{\"a}tter and Sophia Althammer and Michael Schr{\"o}der and Mete Sertkan and Allan Hanbury}, year={2020}, eprint={2010.02666}, archivePrefix={arXiv}, primaryClass={cs.IR} } ```