Parameter-efficient fine-tuning (PEFT) is crucial for customizing Large Language Models (LLMs) with constrained resources.Although there have been various PEFT methods for dense-architecture LLMs, PEFT for sparsearchitecture LLMs is still underexplored.In this work, we study the PEFT method for LLMs with the Mixture-of-Experts (MoE) architecture and the contents of this work are mainly threefold: (1) We investigate the dispersion degree of the activated experts in customized tasks, and found that the routing distribution for a specific task tends to be highly concentrated, while the distribution of activated experts varies significantly across different tasks.(2) We propose Expert-Specialized Fine-Tuning, or ESFT, which tunes the experts most relevant to downstream tasks while freezing the other experts and modules; experimental results demonstrate that our method not only improves the tuning efficiency, but also matches or even surpasses the performance of fullparameter fine-tuning.(3) We further analyze the impact of the MoE architecture on expertspecialized fine-tuning.We find that MoE models with finer-grained experts are more advantageous in selecting the combination of experts that are most relevant to downstream tasks, thereby enhancing both the training efficiency and effectiveness.Our code is available at https://github.com/deepseek-ai/ESFT.

To solve complex tasks, large language models (LLMs) often require multiple rounds of interactions with the user, sometimes assisted by external tools. However, current evaluation protocols often emphasize benchmark performance with single-turn exchanges, neglecting the nuanced interactions among the user, LLMs, and external tools, while also underestimating the importance of natural language feedback from users. These oversights contribute to discrepancies between research benchmark evaluations and real-world use cases. We introduce MINT, a benchmark that evaluates LLMs' ability to solve tasks with multi-turn interactions by (1) using tools and (2) leveraging natural language feedback. To ensure reproducibility, we provide an evaluation framework where LLMs can access tools by executing Python code and receive users' natural language feedback simulated by GPT-4. We repurpose a diverse set of established evaluation datasets focusing on reasoning, coding, and decision-making and carefully curate them into a compact subset for efficient evaluation. Our analysis of 20 open- and closed-source LLMs offers intriguing findings. (a) LLMs generally benefit from tools and language feedback, with performance gains (absolute, same below) of 1--8% for each turn of tool use and 2--17% with natural language feedback. (b) Better single-turn performance does not guarantee better multi-turn performance. (c) Surprisingly, on the LLMs evaluated, supervised instruction-finetuning (SIFT) and reinforcement learning from human feedback (RLHF) generally hurt multi-turn capabilities. We expect MINT can help measure progress and incentivize research in improving LLMs' capabilities in multi-turn interactions, especially for open-source communities where multi-turn human evaluation can be less accessible compared to commercial LLMs with a larger user base.

Since language models (LMs) now outperform average humans on many challenging tasks, it is becoming increasingly difficult to develop challenging, high-quality, and realistic evaluations. We address this by examining LM capabilities to generate code for solving real scientific research problems. Incorporating input from scientists and AI researchers in 16 diverse natural science sub-fields, including mathematics, physics, chemistry, biology, and materials science, we create a scientist-curated coding benchmark, SciCode. The problems naturally factorize into multiple subproblems, each involving knowledge recall, reasoning, and code synthesis. In total, SciCode contains 338 subproblems decomposed from 80 challenging main problems, and it offers optional descriptions specifying useful scientific background information and scientist-annotated gold-standard solutions and test cases for evaluation. OpenAI o1-preview, the best-performing model among those tested, can solve only 7.7\% of the problems in the most realistic setting. We believe that SciCode demonstrates both contemporary LMs' progress towards realizing helpful scientific assistants and sheds light on the building and evaluation of scientific AI in the future.

Model-based Information Retrieval (Model-based IR) has gained attention due to advancements in generative language models. Unlike traditional dense retrieval methods relying on dense vector representations of documents, model-based IR leverages language models to retrieve documents by generating their unique discrete identifiers (docids). This approach effectively reduces the requirements to store separate document representations in an index. Most existing model-based IR approaches utilize pre-defined static docids, i.e., these docids are fixed and are not learnable by training on the retrieval tasks. However, these docids are not specifically optimized for retrieval tasks, which makes it difficult to learn semantics and relationships between documents and achieve satisfactory retrieval performance. To address the above limitations, we propose Neural Optimized VOcabularial (NOVO) docids. NOVO docids are unique n-gram sets identifying each document. They can be generated in any order to retrieve the corresponding document and can be optimized through training to better learn semantics and relationships between documents. We propose to optimize NOVO docids through query denoising modeling and retrieval tasks, allowing for optimizing both semantic and token representations for such docids. Experiments on two datasets under the normal and zero-shot settings show that NOVO exhibits strong performance in more effective and interpretable model-based IR.