Orca 2: Teaching Small Language Models How to Reason

Orca 1 learns from rich signals, such as explanation traces, allowing it to outperform conventional instruction-tuned models on benchmarks like BigBench Hard and AGIEval. In Orca 2, we continue exploring how improved training signals can enhance smaller LMs' reasoning abilities. Research on training small LMs has often relied on imitation learning to replicate the output of more capable models. We contend that excessive emphasis on imitation may restrict the potential of smaller models. We seek to teach small LMs to employ different solution strategies for different tasks, potentially different from the one used by the larger model. For example, while larger models might provide a direct answer to a complex task, smaller models may not have the same capacity. In Orca 2, we teach the model various reasoning techniques (step-by-step, recall then generate, recall-reason-generate, direct answer, etc.). More crucially, we aim to help the model learn to determine the most effective solution strategy for each task. We evaluate Orca 2 using a comprehensive set of 15 diverse benchmarks (corresponding to approximately 100 tasks and over 36,000 unique prompts). Orca 2 significantly surpasses models of similar size and attains performance levels similar or better to those of models 5-10x larger, as assessed on complex tasks that test advanced reasoning abilities in zero-shot settings. make Orca 2 weights publicly available at aka.ms/orca-lm to support research on the development, evaluation, and alignment of smaller LMs

Contrastive Post-training Large Language Models on Data Curriculum

Alignment serves as an important step to steer large language models (LLMs) towards human preferences. In this paper, we explore contrastive post-training techniques for alignment by automatically constructing preference pairs from multiple models of varying strengths (e.g., InstructGPT, ChatGPT and GPT-4). We carefully compare the contrastive techniques of SLiC and DPO to SFT baselines and find that DPO provides a step-function improvement even after continueing SFT saturates. We also explore a data curriculum learning scheme for contrastive post-training, which starts by learning from "easier" pairs and transitioning to "harder" ones, which further improves alignment. Finally, we scale up our experiments to train with more data and larger models like Orca. Remarkably, contrastive post-training further improves the performance of Orca, already a state-of-the-art instruction learning model tuned with GPT-4 outputs, to exceed that of ChatGPT.

SkipDecode: Autoregressive Skip Decoding with Batching and Caching for Efficient LLM Inference

Autoregressive large language models (LLMs) have made remarkable progress in various natural language generation tasks. However, they incur high computation cost and latency resulting from the autoregressive token-by-token generation. To address this issue, several approaches have been proposed to reduce computational cost using early-exit strategies. These strategies enable faster text generation using reduced computation without applying the full computation graph to each token. While existing token-level early exit methods show promising results for online inference, they cannot be readily applied for batch inferencing and Key-Value caching. This is because they have to wait until the last token in a batch exits before they can stop computing. This severely limits the practical application of such techniques. In this paper, we propose a simple and effective token-level early exit method, SkipDecode, designed to work seamlessly with batch inferencing and KV caching. It overcomes prior constraints by setting up a singular exit point for every token in a batch at each sequence position. It also guarantees a monotonic decrease in exit points, thereby eliminating the need to recompute KV Caches for preceding tokens. Rather than terminating computation prematurely as in prior works, our approach bypasses lower to middle layers, devoting most of the computational resources to upper layers, allowing later tokens to benefit from the compute expenditure by earlier tokens. Our experimental results show that SkipDecode can obtain 2x to 5x inference speedups with negligible regression across a variety of tasks. This is achieved using OPT models of 1.3 billion and 6.7 billion parameters, all the while being directly compatible with batching and KV caching optimization techniques.

Unsupervised Extraction of Market Moving Events with Neural Attention

We present a method to identify relevant events associated with stock price movements without manually labeled data. We train an attention-based neural network, which given a set of news headlines for a given time frame, predicts the price movement of a given stock index (i.e., DOWN, STAY, UP). An attention layer acts as an input selector; it computes a normalized weight for each headline embedding. The weighted average of the embeddings is used to predict the price movement. We present an analysis to understand if, after the network has been trained, the attention layer is capable of generating a global ranking of news events through its unnormalized weights. The ranking should be able to rank relevant financial events higher. In this initial study we use news categories as a proxy for relevance: news belonging to more relevant categories should be ranked higher. Our experiments on four indices suggest that there is an indication that the weights indeed skew the global set of events towards those categories that are more relevant to explain the price change; this effect reflects the performance of the network on stock prediction.

KnowNER: Incremental Multilingual Knowledge in Named Entity Recognition

KnowNER is a multilingual Named Entity Recognition (NER) system that leverages different degrees of external knowledge. A novel modular framework divides the knowledge into four categories according to the depth of knowledge they convey. Each category consists of a set of features automatically generated from different information sources (such as a knowledge-base, a list of names or document-specific semantic annotations) and is used to train a conditional random field (CRF). Since those information sources are usually multilingual, KnowNER can be easily trained for a wide range of languages. In this paper, we show that the incorporation of deeper knowledge systematically boosts accuracy and compare KnowNER with state-of-the-art NER approaches across three languages (i.e., English, German and Spanish) performing amongst state-of-the art systems in all of them.