This repository provides pre-trained encoder-decoder models and its related optimization techniques developed by Alibaba's MinD (Machine IntelligeNce of Damo) Lab.
The family of AliceMind:
CVPR 2020 VQA Challenge Runner-up)
StructBERT (March 15, 2021): pre-trained models for natural language understanding (NLU). We extend BERT to a new model, StructBERT, by incorporating language structures into pre-training. Specifically, we pre-train StructBERT with two auxiliary tasks to make the most of the sequential order of words and sentences, which leverage language structures at the word and sentence levels, respectively. "StructBERT: Incorporating Language Structures into Pre-training for Deep Language Understanding" (
PALM (March 15, 2021): pre-trained models for natural language generation (NLG). We propose a novel scheme that jointly pre-trains an autoencoding and autoregressive language model on a large unlabeled corpus, specifically designed for generating new text conditioned on context. It achieves new SOTA results in several downstream tasks. "PALM: Pre-training an Autoencoding&Autoregressive Language Model for Context-conditioned Generation" (
VECO v0 (March 15, 2021): pre-trained models for cross-lingual (x) natural language understanding (x-NLU) and generation (x-NLG). VECO (v0) achieves the new SOTA results on various cross-lingual understanding tasks of the XTREME benchmark, covering text classification, sequence labeling, question answering, and sentence retrieval. For cross-lingual generation tasks, it also outperforms all existing cross-lingual models and state-of-the-art Transformer variants on WMT14 English-to-German and English-to-French translation datasets, with gains of up to 1~2 BLEU. “VECO: Variable Encoder-decoder Pre-training for Cross-lingual Understanding and Generation" (
StructVBERT (March 15, 2021): pre-trained models for vision-language understanding. We propose a new single-stream visual-linguistic pre-training scheme by leveraging multi-stage progressive pre-training and multi-task learning. StructVBERT obtained the 2020 VQA Challenge Runner-up award, and SOTA result on VQA 2020 public Test-standard benchmark (June 2020). "Talk Slides" (
CVPR 2020 VQA Challenge Runner-up).
StructuralLM (March 15, 2021): pre-trained models for document-image understanding. We propose a new pre-training approach, StructuralLM, to jointly leverage cell and layout information from scanned documents. The pre-trained StructuralLM achieves new state-of-the-art results in different types of downstream tasks. "StructuralLM: Structural Pre-training for Form Understanding" (
LatticeBERT (March 15, 2021): we propose a novel pre-training paradigm for Chinese — Lattice-BERT which explicitly incorporates word representations with those of characters, thus can model a sentence in a multi-granularity manner. "Lattice-BERT: Leveraging Multi-Granularity Representations in Chinese Pre-trained Language Models" (
SDCUP (September 6, 2021): pre-trained models for table understanding. We design a schema dependency pre-training objective to impose the desired inductive bias into the learned representations for table pre-training. We further propose a schema-aware curriculum learning approach to alleviate the impact of noise and learn effectively from the pre-training data in an easy-to-hard manner. The experiment results on SQUALL and Spider demonstrate the effectiveness of our pre-training objective and curriculum in comparison to a variety of baselines. "SDCUP: Schema Dependency Enhanced Curriculum Pre-Training for Table Semantic Parsing" (
PLUG (September 1, 2022): large-scale chinese pre-trained model for understanding and generation. PLUG (27B) is a large-scale chinese pre-training model for language understanding and generation. The training of PLUG is two-stage, the first stage is a 24-layer StructBERT encoder, and the second stage is a 24-6-layer PALM encoder-decoder.
mPLUG (September 1, 2022): large-scale pre-trained model for vision-language understanding and generation. mPLUG is pre-trained end-to-end on large scale image-text pairs with both discriminative and generative objectives. It achieves state-of-the-art results on a wide range of vision-language downstream tasks, including image-captioning, image-text retrieval, visual grounding and visual question answering.
ContrastivePruning (December 17, 2021):
ContrAstive Pruning (CAP) is a general pruning framework under the pre-training and fine-tuning paradigm, which aims at maintaining both task-specific and task-agnostic knowledge during pruning. CAP is designed as a general framework, compatible with both structured and unstructured pruning. Unified in contrastive learning, CAP encourage the pruned model to learn from the pre-trained model, the snapshots (intermediate models during pruning), and the fine-tuned model, respectively. “From Dense to Sparse: Contrastive Pruning for Better Pre-trained Language Model Compression" (
PST (May 23, 2022):
Parameter-efficient Sparse Training (PST) is to reduce the number of trainable parameters during sparse-aware training in downstream tasks. It combines the data-free and data-driven criteria to efficiently and accurately measures the importance of weights, and investigates the intrinsic redundancy of data-driven weight importance and derive two obvious characteristics i.e., low-rankness and structuredness, which therefore makes the sparse training resource-efficient and parameter-efficient. “Parameter-Efficient Sparsity for Large Language Models Fine-Tuning" (
AliceMind Official Website: https://nlp.aliyun.com/portal#/alice
AliceMind Open Platform: https://alicemind.aliyuncs.com
Please submit a GitHub issue if you have want help or have issues using ALICE.
For more information, you can join the
AliceMind Users Group on DingTalk to contact us. The number of the DingTalk group is 35738533.
For other business communications, please contact [email protected]
AliceMind is released under the Apache 2.0 license.
Copyright 1999-2020 Alibaba Group Holding Ltd. Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at the following link. http://www.apache.org/licenses/LICENSE-2.0 Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.