Large Language Models (LLMs) can correct their self-generated responses, but a decline in accuracy after self-correction is also witnessed. To have a deeper understanding of self-correction, we endeavor to decompose, evaluate, and analyze the self-correction behaviors of LLMs. By enumerating and analyzing answer correctness before and after self-correction, we decompose the self-correction capability into confidence (being confident to correct answers) and critique (turning wrong answers to correct) capabilities, and propose two metrics from a probabilistic perspective to measure these 2 capabilities, along with another metric for overall self-correction capability evaluation. Based on our decomposition and evaluation metrics, we conduct extensive experiments and draw some empirical conclusions. For example, we find different models can exhibit distinct behaviors: some models are confident while others are more critical. We also find the trade-off between the two capabilities (i.e. improving one can lead to a decline in the other) when manipulating model self-correction behavior by prompts or in-context learning. Further, we find a simple yet efficient strategy to improve self-correction capability by transforming Supervision Fine-Tuning (SFT) data format, and our strategy outperforms vanilla SFT in both capabilities and achieves much higher accuracy after self-correction. Our code will be publicly available on GitHub.

Open-source Large Language Models (LLMs) and their specialized variants, particularly Code LLMs, have recently delivered impressive performance. However, previous Code LLMs are typically fine-tuned on single-source data with limited quality and diversity, which may insufficiently elicit the potential of pre-trained Code LLMs. In this paper, we present AlchemistCoder, a series of Code LLMs with enhanced code generation and generalization capabilities fine-tuned on multi-source data. To achieve this, we pioneer to unveil inherent conflicts among the various styles and qualities in multi-source code corpora and introduce data-specific prompts with hindsight relabeling, termed AlchemistPrompts, to harmonize different data sources and instruction-response pairs. Additionally, we propose incorporating the data construction process into the fine-tuning data as code comprehension tasks, including instruction evolution, data filtering, and code review. Extensive experiments demonstrate that AlchemistCoder holds a clear lead among all models of the same size (6.7B/7B) and rivals or even surpasses larger models (15B/33B/70B), showcasing the efficacy of our method in refining instruction-following capabilities and advancing the boundaries of code intelligence.

Most work treats large language models as black boxes without in-depth understanding of their internal working mechanism. In order to explain the internal representations of LLMs, we propose a gradient-based metric to assess the activation level of model parameters. Based on this metric, we obtain three preliminary findings. (1) When the inputs are in the same domain, parameters in the shallow layers will be activated densely, which means a larger portion of parameters will have great impacts on the outputs. In contrast, parameters in the deep layers are activated sparsely. (2) When the inputs are across different domains, parameters in shallow layers exhibit higher similarity in the activation behavior than deep layers. (3) In deep layers, the similarity of the distributions of activated parameters is positively correlated to the empirical data relevance. Further, we develop three validation experiments to solidify these findings. (1) Firstly, starting from the first finding, we attempt to configure different prune ratios for different layers, and find this method can benefit model pruning. (2) Secondly, we find that a pruned model based on one calibration set can better handle tasks related to the calibration task than those not related, which validate the second finding. (3) Thirdly, Based on the STS-B and SICK benchmark, we find that two sentences with consistent semantics tend to share similar parameter activation patterns in deep layers, which aligns with our third finding. Our work sheds light on the behavior of parameter activation in LLMs, and we hope these findings will have the potential to inspire more practical applications.

The evolution of Large Language Models (LLMs) like ChatGPT and GPT-4 has sparked discussions on the advent of Artificial General Intelligence (AGI). However, replicating such advancements in open-source models has been challenging. This paper introduces InternLM2, an open-source LLM that outperforms its predecessors in comprehensive evaluations across 6 dimensions and 30 benchmarks, long-context modeling, and open-ended subjective evaluations through innovative pre-training and optimization techniques. The pre-training process of InternLM2 is meticulously detailed, highlighting the preparation of diverse data types including text, code, and long-context data. InternLM2 efficiently captures long-term dependencies, initially trained on 4k tokens before advancing to 32k tokens in pre-training and fine-tuning stages, exhibiting remarkable performance on the 200k ``Needle-in-a-Haystack" test. InternLM2 is further aligned using Supervised Fine-Tuning (SFT) and a novel Conditional Online Reinforcement Learning from Human Feedback (COOL RLHF) strategy that addresses conflicting human preferences and reward hacking. By releasing InternLM2 models in different training stages and model sizes, we provide the community with insights into the model's evolution.

The programming skill is one crucial ability for Large Language Models (LLMs), necessitating a deep understanding of programming languages (PLs) and their correlation with natural languages (NLs). We examine the impact of pre-training data on code-focused LLMs' performance by assessing the comment density as a measure of PL-NL alignment. Given the scarcity of code-comment aligned data in pre-training corpora, we introduce a novel data augmentation method that generates comments for existing code, coupled with a data filtering strategy that filters out code data poorly correlated with natural language. We conducted experiments on three code-focused LLMs and observed consistent improvements in performance on two widely-used programming skill benchmarks. Notably, the model trained on the augmented data outperformed both the model used for generating comments and the model further trained on the data without augmentation.

The math abilities of large language models can represent their abstract reasoning ability. In this paper, we introduce and open-source our math reasoning LLMs InternLM-Math which is continue pre-trained from InternLM2. We unify chain-of-thought reasoning, reward modeling, formal reasoning, data augmentation, and code interpreter in a unified seq2seq format and supervise our model to be a versatile math reasoner, verifier, prover, and augmenter. These abilities can be used to develop the next math LLMs or self-iteration. InternLM-Math obtains open-sourced state-of-the-art performance under the setting of in-context learning, supervised fine-tuning, and code-assisted reasoning in various informal and formal benchmarks including GSM8K, MATH, Hungary math exam, MathBench-ZH, and MiniF2F. Our pre-trained model achieves 30.3 on the MiniF2F test set without fine-tuning. We further explore how to use LEAN to solve math problems and study its performance under the setting of multi-task learning which shows the possibility of using LEAN as a unified platform for solving and proving in math. Our models, codes, and data are released at \url{https://github.com/InternLM/InternLM-Math}.

We present a vision and language model named MultiModal-GPT to conduct multi-round dialogue with humans. MultiModal-GPT can follow various instructions from humans, such as generating a detailed caption, counting the number of interested objects, and answering general questions from users. MultiModal-GPT is parameter-efficiently fine-tuned from OpenFlamingo, with Low-rank Adapter (LoRA) added both in the cross-attention part and the self-attention part of the language model. We first construct instruction templates with vision and language data for multi-modality instruction tuning to make the model understand and follow human instructions. We find the quality of training data is vital for the dialogue performance, where few data containing short answers can lead the model to respond shortly to any instructions. To further enhance the ability to chat with humans of the MultiModal-GPT, we utilize language-only instruction-following data to train the MultiModal-GPT jointly. The joint training of language-only and visual-language instructions with the \emph{same} instruction template effectively improves dialogue performance. Various demos show the ability of continuous dialogue of MultiModal-GPT with humans. Code, dataset, and demo are at https://github.com/open-mmlab/Multimodal-GPT

With promising yet saturated results in high-resource settings, low-resource datasets have gradually become popular benchmarks for evaluating the learning ability of advanced neural networks (e.g., BigBench, superGLUE). Some models even surpass humans according to benchmark test results. However, we find that there exists a set of hard examples in low-resource settings that challenge neural networks but are not well evaluated, which causes over-estimated performance. We first give a theoretical analysis on which factors bring the difficulty of low-resource learning. It then motivate us to propose a challenging benchmark hardBench to better evaluate the learning ability, which covers 11 datasets, including 3 computer vision (CV) datasets and 8 natural language process (NLP) datasets. Experiments on a wide range of models show that neural networks, even pre-trained language models, have sharp performance drops on our benchmark, demonstrating the effectiveness on evaluating the weaknesses of neural networks. On NLP tasks, we surprisingly find that despite better results on traditional low-resource benchmarks, pre-trained networks, does not show performance improvements on our benchmarks. These results demonstrate that there are still a large robustness gap between existing models and human-level performance.