Being able to thoroughly assess massive multi-task language understanding (MMLU) capabilities is essential for advancing the applicability of multilingual language models. However, preparing such benchmarks in high quality native language is often costly and therefore limits the representativeness of evaluation datasets. While recent efforts focused on building more inclusive MMLU benchmarks, these are conventionally built using machine translation from high-resource languages, which may introduce errors and fail to account for the linguistic and cultural intricacies of the target languages. In this paper, we address the lack of native language MMLU benchmark especially in the under-represented Turkic language family with distinct morphosyntactic and cultural characteristics. We propose two benchmarks for Turkic language MMLU: TUMLU is a comprehensive, multilingual, and natively developed language understanding benchmark specifically designed for Turkic languages. It consists of middle- and high-school level questions spanning 11 academic subjects in Azerbaijani, Crimean Tatar, Karakalpak, Kazakh, Tatar, Turkish, Uyghur, and Uzbek. We also present TUMLU-mini, a more concise, balanced, and manually verified subset of the dataset. Using this dataset, we systematically evaluate a diverse range of open and proprietary multilingual large language models (LLMs), including Claude, Gemini, GPT, and LLaMA, offering an in-depth analysis of their performance across different languages, subjects, and alphabets. To promote further research and development in multilingual language understanding, we release TUMLU-mini and all corresponding evaluation scripts.
We introduce our first-generation reasoning models, DeepSeek-R1-Zero and DeepSeek-R1. DeepSeek-R1-Zero, a model trained via large-scale reinforcement learning (RL) without supervised fine-tuning (SFT) as a preliminary step, demonstrates remarkable reasoning capabilities. Through RL, DeepSeek-R1-Zero naturally emerges with numerous powerful and intriguing reasoning behaviors. However, it encounters challenges such as poor readability, and language mixing. To address these issues and further enhance reasoning performance, we introduce DeepSeek-R1, which incorporates multi-stage training and cold-start data before RL. DeepSeek-R1 achieves performance comparable to OpenAI-o1-1217 on reasoning tasks. To support the research community, we open-source DeepSeek-R1-Zero, DeepSeek-R1, and six dense models (1.5B, 7B, 8B, 14B, 32B, 70B) distilled from DeepSeek-R1 based on Qwen and Llama.
Multiple-choice question (MCQ) datasets like Massive Multitask Language Understanding (MMLU) are widely used to evaluate the commonsense, understanding, and problem-solving abilities of large language models (LLMs). However, the open-source nature of these benchmarks and the broad sources of training data for LLMs have inevitably led to benchmark contamination, resulting in unreliable evaluation results. To alleviate this issue, we propose a contamination-free and more challenging MCQ benchmark called MMLU-CF. This benchmark reassesses LLMs' understanding of world knowledge by averting both unintentional and malicious data leakage. To avoid unintentional data leakage, we source data from a broader domain and design three decontamination rules. To prevent malicious data leakage, we divide the benchmark into validation and test sets with similar difficulty and subject distributions. The test set remains closed-source to ensure reliable results, while the validation set is publicly available to promote transparency and facilitate independent verification. Our evaluation of mainstream LLMs reveals that the powerful GPT-4o achieves merely a 5-shot score of 73.4% and a 0-shot score of 71.9% on the test set, which indicates the effectiveness of our approach in creating a more rigorous and contamination-free evaluation standard. The GitHub repository is available at https://github.com/microsoft/MMLU-CF and the dataset refers to https://huggingface.co/datasets/microsoft/MMLU-CF.
Scaling large language models (LLMs) significantly improves performance but comes with prohibitive computational costs. Mixture-of-Experts (MoE) models offer an efficient alternative, increasing capacity without a proportional rise in compute requirements. However, training MoE models from scratch poses challenges like overfitting and routing instability. We present an efficient training recipe leveraging pre-trained dense checkpoints, training an 8-Expert Top-2 MoE model from Llama 3-8B with less than $1\%$ of typical pre-training compute. Our approach enhances downstream performance on academic benchmarks, achieving a $\textbf{2%}$ improvement in 0-shot accuracy on MMLU, while reaching a Model FLOPs Utilization (MFU) of $\textbf{46.8%}$ during training using our framework. We also integrate online upcycling in NeMo for seamless use of pre-trained weights, enabling cost-effective development of high-capacity MoE models.
Assessing the effectiveness of large language models (LLMs) in performing different tasks is crucial for understanding their strengths and weaknesses. This paper presents Hierarchical Prompting Taxonomy (HPT), grounded on human cognitive principles and designed to assess LLMs by examining the cognitive demands of various tasks. The HPT utilizes the Hierarchical Prompting Framework (HPF), which structures five unique prompting strategies in a hierarchical order based on their cognitive requirement on LLMs when compared to human mental capabilities. It assesses the complexity of tasks with the Hierarchical Prompting Index (HPI), which demonstrates the cognitive competencies of LLMs across diverse datasets and offers insights into the cognitive demands that datasets place on different LLMs. This approach enables a comprehensive evaluation of an LLMs problem solving abilities and the intricacy of a dataset, offering a standardized metric for task complexity. Extensive experiments with multiple datasets and LLMs show that HPF enhances LLM performance by 2% to 63% compared to baseline performance, with GSM8k being the most cognitively complex task among reasoning and coding tasks with an average HPI of 3.20 confirming the effectiveness of HPT. To support future research and reproducibility in this domain, the implementations of HPT and HPF are available here.
Ensembling multiple models has always been an effective approach to push the limits of existing performance and is widely used in classification tasks by simply averaging the classification probability vectors from multiple classifiers to achieve better accuracy. However, in the thriving open-source Large Language Model (LLM) community, ensembling methods are rare and typically limited to ensembling the full-text outputs of LLMs, such as selecting the best output using a ranker, which leads to underutilization of token-level probability information. In this paper, we treat the Generation of each token by LLMs as a Classification (GaC) for ensembling. This approach fully exploits the probability information at each generation step and better prevents LLMs from producing early incorrect tokens that lead to snowballing errors. In experiments, we ensemble state-of-the-art LLMs on several benchmarks, including exams, mathematics and reasoning, and observe that our method breaks the existing community performance ceiling. Furthermore, we observed that most of the tokens in the answer are simple and do not affect the correctness of the final answer. Therefore, we also experimented with ensembling only key tokens, and the results showed better performance with lower latency across benchmarks.
The focus of language model evaluation has transitioned towards reasoning and knowledge-intensive tasks, driven by advancements in pretraining large models. While state-of-the-art models are partially trained on large Arabic texts, evaluating their performance in Arabic remains challenging due to the limited availability of relevant datasets. To bridge this gap, we present \datasetname{}, the first multi-task language understanding benchmark for the Arabic language, sourced from school exams across diverse educational levels in different countries spanning North Africa, the Levant, and the Gulf regions. Our data comprises 40 tasks and 14,575 multiple-choice questions in Modern Standard Arabic (MSA) and is carefully constructed by collaborating with native speakers in the region. Our comprehensive evaluations of 35 models reveal substantial room for improvement, particularly among the best open-source models. Notably, BLOOMZ, mT0, LLaMA2, and Falcon struggle to achieve a score of 50%, while even the top-performing Arabic-centric model only achieves a score of 62.3%.
LLMs with superior response quality--particularly larger or closed-source models--often come with higher inference costs, making their deployment inefficient and costly. Meanwhile, developing foundational LLMs from scratch is becoming increasingly resource-intensive and impractical for many applications. To address the challenge of balancing quality and cost, we introduce Routoo, an architecture designed to optimize the selection of LLMs for specific prompts based on performance, cost, and efficiency. Routoo provides controllability over the trade-off between inference cost and quality, enabling significant reductions in inference costs for a given quality requirement. Routoo comprises two key components: a performance predictor and cost-aware selector. The performance predictor is a lightweight LLM that estimates the expected performance of various underlying LLMs on a given prompt without executing them. The cost-aware selector module then selects the most suitable model based on these predictions and constraints such as cost and latency, significantly reducing inference costs for the same quality. We evaluated Routoo using the MMLU benchmark across 57 domains employing open-source models. Our results show that Routoo matches the performance of the Mixtral 8x7b model while reducing inference costs by one-third. Additionally, by allowing increased costs, Routoo surpasses Mixtral's accuracy by over 5% at equivalent costs, achieving an accuracy of 75.9%. When integrating GPT4 into our model pool, Routoo nearly matches GPT4's performance at half the cost and exceeds it with a 25% cost reduction. These outcomes highlight Routoo's potential to significantly reduce inference costs without compromising quality, and even to establish new state-of-the-art results by leveraging the collective capabilities of multiple LLMs.
Large language models (LLMs) have demonstrated considerable proficiency in general natural language processing (NLP) tasks. Instruction tuning, a successful paradigm, enhances the ability of LLMs to follow natural language instructions and exhibit robust generalization across general tasks. However, these models often encounter performance limitations across multiple tasks due to constrained model capacity. Expanding this capacity during the instruction tuning phase poses significant challenges. To address this issue, we introduce parameter-efficient sparsity crafting (PESC), which crafts dense models into sparse models using the mixture-of-experts (MoE) architecture. PESC integrates adapters into the MoE layers of sparse models, differentiating experts without altering the individual weights within these layers. This method significantly reduces computational costs and GPU memory requirements, facilitating model capacity expansion through a minimal parameter increase when guaranteeing the quality of approximation in function space compared to original sparse upcycling. Our empirical evaluation demonstrates the effectiveness of the PESC method. Using PESC during instruction tuning, our best sparse model outperforms other sparse and dense models and exhibits superior general capabilities compared to GPT-3.5. Our code is available at https://github.com/wuhy68/Parameter-Efficient-MoE.
Generative language models are usually pretrained on large text corpus via predicting the next token (i.e., sub-word/word/phrase) given the previous ones. Recent works have demonstrated the impressive performance of large generative language models on downstream tasks. However, existing generative language models generally neglect an inherent challenge in text corpus during training, i.e., the imbalance between frequent tokens and infrequent ones. It can lead a language model to be dominated by common and easy-to-learn tokens, thereby overlooking the infrequent and difficult-to-learn ones. To alleviate that, we propose a MiLe Loss function for mitigating the bias of learning difficulties with tokens. During training, it can dynamically assess the learning difficulty of a to-be-learned token, according to the information entropy of the corresponding predicted probability distribution over the vocabulary. Then it scales the training loss adaptively, trying to lead the model to focus more on the difficult-to-learn tokens. On the Pile dataset, we train generative language models at different scales of 468M, 1.2B, and 6.7B parameters. Experiments reveal that models incorporating the proposed MiLe Loss can gain consistent performance improvement on downstream benchmarks.
We introduce Mistral 7B v0.1, a 7-billion-parameter language model engineered for superior performance and efficiency. Mistral 7B outperforms Llama 2 13B across all evaluated benchmarks, and Llama 1 34B in reasoning, mathematics, and code generation. Our model leverages grouped-query attention (GQA) for faster inference, coupled with sliding window attention (SWA) to effectively handle sequences of arbitrary length with a reduced inference cost. We also provide a model fine-tuned to follow instructions, Mistral 7B -- Instruct, that surpasses the Llama 2 13B -- Chat model both on human and automated benchmarks. Our models are released under the Apache 2.0 license.
Although large language models (LLMs) are often pre-trained on large-scale multilingual texts, their reasoning abilities and real-world knowledge are mainly evaluated based on English datasets. Assessing LLM capabilities beyond English is increasingly vital but hindered due to the lack of suitable datasets. In this work, we introduce IndoMMLU, the first multi-task language understanding benchmark for Indonesian culture and languages, which consists of questions from primary school to university entrance exams in Indonesia. By employing professional teachers, we obtain 14,981 questions across 64 tasks and education levels, with 46% of the questions focusing on assessing proficiency in the Indonesian language and knowledge of nine local languages and cultures in Indonesia. Our empirical evaluations show that GPT-3.5 only manages to pass the Indonesian primary school level, with limited knowledge of local Indonesian languages and culture. Other smaller models such as BLOOMZ and Falcon perform at even lower levels.
Despite the promising few-shot ability of large language models (LLMs), the standard paradigm of In-context Learning (ICL) suffers the disadvantages of susceptibility to selected demonstrations and the intricacy to generate these demonstrations. In this paper, we raise the fundamental question that whether human-generated demonstrations are necessary for ICL. To answer this question, we propose self-contemplation prompting strategy (SEC), a paradigm free from human-crafted demonstrations. The key point of SEC is that, instead of using hand-crafted examples as demonstrations in ICL, SEC asks LLMs to first create demonstrations on their own, based on which the final output is generated. SEC is a flexible framework and can be adapted to both the vanilla ICL and the chain-of-thought (CoT), but with greater ease: as the manual-generation process of both examples and rationale can be saved. Extensive experiments in arithmetic reasoning, commonsense reasoning, multi-task language understanding, and code generation benchmarks, show that SEC, which does not require hand-crafted demonstrations, significantly outperforms the zero-shot learning strategy, and achieves comparable results to ICL with hand-crafted demonstrations. This demonstrates that, for many tasks, contemporary LLMs possess a sufficient level of competence to exclusively depend on their own capacity for decision making, removing the need for external training data. Code is available at https://github.com/ruili33/SEC.
In this work, we develop and release Llama 2, a collection of pretrained and fine-tuned large language models (LLMs) ranging in scale from 7 billion to 70 billion parameters. Our fine-tuned LLMs, called Llama 2-Chat, are optimized for dialogue use cases. Our models outperform open-source chat models on most benchmarks we tested, and based on our human evaluations for helpfulness and safety, may be a suitable substitute for closed-source models. We provide a detailed description of our approach to fine-tuning and safety improvements of Llama 2-Chat in order to enable the community to build on our work and contribute to the responsible development of LLMs.
The drastic increase in language models' parameters has led to a new trend of deploying models in cloud servers, raising growing concerns about private inference for Transformer-based models. Existing two-party privacy-preserving techniques, however, only take into account natural language understanding (NLU) scenarios. Private inference in natural language generation (NLG), crucial for applications like translation and code completion, remains underexplored.In addition, previous privacy-preserving techniques suffer from convergence issues during model training and exhibit poor inference speed when used with NLG models due to the neglect of time-consuming operations in auto-regressive generations. To address these issues, we propose a fast private text generation framework for Transformer-based language models, namely MERGE.MERGE reuses the output hidden state as the word embedding to bypass the embedding computation and reorganize the linear operations in the Transformer module to accelerate the forward procedure. Extensive experiments show that MERGE achieves a 26.5x speedup to the vanilla encrypted model under the sequence length 512, and reduces 80\% communication cost, with an up to 10x speedup to state-of-the-art approximated models.
We report the development of GPT-4, a large-scale, multimodal model which can accept image and text inputs and produce text outputs. While less capable than humans in many real-world scenarios, GPT-4 exhibits human-level performance on various professional and academic benchmarks, including passing a simulated bar exam with a score around the top 10% of test takers. GPT-4 is a Transformer-based model pre-trained to predict the next token in a document. The post-training alignment process results in improved performance on measures of factuality and adherence to desired behavior. A core component of this project was developing infrastructure and optimization methods that behave predictably across a wide range of scales. This allowed us to accurately predict some aspects of GPT-4's performance based on models trained with no more than 1/1,000th the compute of GPT-4.
We introduce LLaMA, a collection of foundation language models ranging from 7B to 65B parameters. We train our models on trillions of tokens, and show that it is possible to train state-of-the-art models using publicly available datasets exclusively, without resorting to proprietary and inaccessible datasets. In particular, LLaMA-13B outperforms GPT-3 (175B) on most benchmarks, and LLaMA-65B is competitive with the best models, Chinchilla-70B and PaLM-540B. We release all our models to the research community.
Information overload is a major obstacle to scientific progress. The explosive growth in scientific literature and data has made it ever harder to discover useful insights in a large mass of information. Today scientific knowledge is accessed through search engines, but they are unable to organize scientific knowledge alone. In this paper we introduce Galactica: a large language model that can store, combine and reason about scientific knowledge. We train on a large scientific corpus of papers, reference material, knowledge bases and many other sources. We outperform existing models on a range of scientific tasks. On technical knowledge probes such as LaTeX equations, Galactica outperforms the latest GPT-3 by 68.2% versus 49.0%. Galactica also performs well on reasoning, outperforming Chinchilla on mathematical MMLU by 41.3% to 35.7%, and PaLM 540B on MATH with a score of 20.4% versus 8.8%. It also sets a new state-of-the-art on downstream tasks such as PubMedQA and MedMCQA dev of 77.6% and 52.9%. And despite not being trained on a general corpus, Galactica outperforms BLOOM and OPT-175B on BIG-bench. We believe these results demonstrate the potential for language models as a new interface for science. We open source the model for the benefit of the scientific community.
We introduce GLM-130B, a bilingual (English and Chinese) pre-trained language model with 130 billion parameters. It is an attempt to open-source a 100B-scale model at least as good as GPT-3 (davinci) and unveil how models of such a scale can be successfully pre-trained. Over the course of this effort, we face numerous unexpected technical and engineering challenges, particularly on loss spikes and divergence. In this paper, we introduce the training process of GLM-130B including its design choices, training strategies for both efficiency and stability, and engineering efforts. The resultant GLM-130B model offers significant outperformance over GPT-3 175B (davinci) on a wide range of popular English benchmarks while the performance advantage is not observed in OPT-175B and BLOOM-176B. It also consistently and significantly outperforms ERNIE TITAN 3.0 260B -- the largest Chinese language model -- across related benchmarks. Finally, we leverage a unique scaling property of GLM-130B to reach INT4 quantization without post training, with almost no performance loss, making it the first among 100B-scale models and more importantly, allowing its effective inference on 4$\times$RTX 3090 (24G) or 8$\times$RTX 2080 Ti (11G) GPUs, the most affordable GPUs required for using 100B-scale models. The GLM-130B model weights are publicly accessible and its code, training logs, related toolkit, and lessons learned are open-sourced at \url{https://github.com/THUDM/GLM-130B/}.
Large language models have shown impressive few-shot results on a wide range of tasks. However, when knowledge is key for such results, as is the case for tasks such as question answering and fact checking, massive parameter counts to store knowledge seem to be needed. Retrieval augmented models are known to excel at knowledge intensive tasks without the need for as many parameters, but it is unclear whether they work in few-shot settings. In this work we present Atlas, a carefully designed and pre-trained retrieval augmented language model able to learn knowledge intensive tasks with very few training examples. We perform evaluations on a wide range of tasks, including MMLU, KILT and NaturalQuestions, and study the impact of the content of the document index, showing that it can easily be updated. Notably, Atlas reaches over 42% accuracy on Natural Questions using only 64 examples, outperforming a 540B parameters model by 3% despite having 50x fewer parameters.