Search

We introduce ChatGLM, an evolving family of large language models that we have been developing over time. This report primarily focuses on the GLM-4 language series, which includes GLM-4, GLM-4-Air, and GLM-4-9B. They represent our most capable models that are trained with all the insights and lessons gained from the preceding three generations of ChatGLM. To date, the GLM-4 models are pre-trained on ten trillions of tokens mostly in Chinese and English, along with a small set of corpus from 24 languages, and aligned primarily for Chinese and English usage. The high-quality alignment is achieved via a multi-stage post-training process, which involves supervised fine-tuning and learning from human feedback. Evaluations show that GLM-4 1) closely rivals or outperforms GPT-4 in terms of general metrics such as MMLU, GSM8K, MATH, BBH, GPQA, and HumanEval, 2) gets close to GPT-4-Turbo in instruction following as measured by IFEval, 3) matches GPT-4 Turbo (128K) and Claude 3 for long context tasks, and 4) outperforms GPT-4 in Chinese alignments as measured by AlignBench. The GLM-4 All Tools model is further aligned to understand user intent and autonomously decide when and which tool(s) touse -- including web browser, Python interpreter, text-to-image model, and user-defined functions -- to effectively complete complex tasks. In practical applications, it matches and even surpasses GPT-4 All Tools in tasks like accessing online information via web browsing and solving math problems using Python interpreter. Over the course, we have open-sourced a series of models, including ChatGLM-6B (three generations), GLM-4-9B (128K, 1M), GLM-4V-9B, WebGLM, and CodeGeeX, attracting over 10 million downloads on Hugging face in the year 2023 alone. The open models can be accessed through https://github.com/THUDM and https://huggingface.co/THUDM.

Large language models (LLMs) have shown excellent mastering of human language, but still struggle in real-world applications that require mathematical problem-solving. While many strategies and datasets to enhance LLMs' mathematics are developed, it remains a challenge to simultaneously maintain and improve both language and mathematical capabilities in deployed LLM systems.In this work, we tailor the Self-Critique pipeline, which addresses the challenge in the feedback learning stage of LLM alignment. We first train a general Math-Critique model from the LLM itself to provide feedback signals. Then, we sequentially employ rejective fine-tuning and direct preference optimization over the LLM's own generations for data collection. Based on ChatGLM3-32B, we conduct a series of experiments on both academic and our newly created challenging dataset, MathUserEval. Results show that our pipeline significantly enhances the LLM's mathematical problem-solving while still improving its language ability, outperforming LLMs that could be two times larger. Related techniques have been deployed to ChatGLM\footnote{\url{https://chatglm.cn}}, an online serving LLM. Related evaluation dataset and scripts are released at \url{https://github.com/THUDM/ChatGLM-Math}.

Recent experiment has uncovered semimetal bismuth (Bi) as an excellent electrical contact to monolayer MoS$_2$ with ultralow contact resistance. The contact physics of the broader semimetal/monolayer-semiconductor family beyond Bi/MoS$_2$, however, remains largely unexplored thus far. Here we perform a comprehensive first-principle density functional theory investigation on the electrical contact properties between six archetypal two-dimensional (2D) transition metal dichalcogenide (TMDC) semiconductors, i.e. MoS$_2$, WS$_2$, MoSe$_2$, WSe$_2$, MoTe$_2$ and WTe$_2$, and two representative types of semimetals, Bi and antimony (Sb). As Bi and Sb work functions energetically aligns well with the TMDC conduction band edge, Ohmic or nearly-Ohmic $n$-type contacts are prevalent. The interlayer distance of semimetal/TMDC contacts are significantly larger than that of the metal/TMDC counterparts, which results in only weak metalization of TMDC upon contact formation. Intriguingly, such weak metalization generates semimetal-induced gap states (MIGS) that extends below the conduction band minimum, thus offering an effective mechanism to reduce or eliminate the $n$-type Schottky barrier height (SBH) while still preserving the electronic structures of 2D TMDC. A modified Schottky-Mott rule that takes into account SMIGS, interface dipole potential, and Fermi level shifting is proposed, which provides an improved agreement with the DFT-simulated SBH. We further show that the tunneling-specific resistivity of Sb/TMDC contacts are generally lower than the Bi counterparts, thus indicating a better charge injection efficiency can be achieved through Sb contacts. Our findings reveal the promising potential of Bi and Sb as excellent companion electrode materials for advancing 2D semiconductor device technology., Comment: 12 pages, 7 figures