Qingfu Zhu

Zhiyuan Ma, Jiayu Liu, Xianzhen Luo et al.

Empowering large language models (LLMs) with effective tool utilization capabilities is crucial for enabling AI agents to solve complex problems. However, current models face two major limitations: (1) unreliable tool planning and invocation due to low-quality instruction datasets (e.g., widespread hallucinated API calls), and (2) weak tool reflection abilities (over 90% of errors cannot be corrected) resulting from static imitation learning. To address these critical limitations, we propose Tool-MVR, a novel Tool-Augmented LLM that achieves comprehensive System 2 reasoning through two key innovations. Specifically, we first introduce Multi-Agent Meta-Verification (MAMV), a systematic pipeline that rigorously validates APIs, queries, and reasoning trajectories to construct ToolBench-V, a new high-quality instruction dataset that addresses the limitation of unreliable tool planning and invocation. Second, we propose Exploration-based Reflection Learning (EXPLORE), which enhances tool reflection capabilities by leveraging tool feedback through a dynamic "Error -> Reflection -> Correction" learning paradigm, resulting in our reflection dataset ToolBench-R and addressing the critical weakness in tool reflection. Finally, we obtain Tool-MVR by finetuning open-source LLMs (e.g., Qwen-7B) on both ToolBench-V and ToolBench-R. Our experiments demonstrate that Tool-MVR achieves state-of-the-art performance on StableToolBench, surpassing both ToolLLM (by 23.9%) and GPT-4 (by 15.3%) while reducing API calls by 31.4%, with strong generalization capabilities across unseen tools and scenarios. Additionally, on our proposed RefineToolBench, the first benchmark specifically designed to evaluate tool reflection capabilities, Tool-MVR achieves a 58.9% error correction rate, significantly outperforming ToolLLM's 9.1%.

Wei-Nan Zhang, Yiming Cui, Kaiyan Zhang et al.

Recently, research on open domain dialogue systems have attracted extensive interests of academic and industrial researchers. The goal of an open domain dialogue system is to imitate humans in conversations. Previous works on single turn conversation generation have greatly promoted the research of open domain dialogue systems. However, understanding multiple single turn conversations is not equal to the understanding of multi turn dialogue due to the coherent and context dependent properties of human dialogue. Therefore, in open domain multi turn dialogue generation, it is essential to modeling the contextual semantics of the dialogue history, rather than only according to the last utterance. Previous research had verified the effectiveness of the hierarchical recurrent encoder-decoder framework on open domain multi turn dialogue generation. However, using RNN-based model to hierarchically encoding the utterances to obtain the representation of dialogue history still face the problem of a vanishing gradient. To address this issue, in this paper, we proposed a static and dynamic attention-based approach to model the dialogue history and then generate open domain multi turn dialogue responses. Experimental results on Ubuntu and Opensubtitles datasets verify the effectiveness of the proposed static and dynamic attention-based approach on automatic and human evaluation metrics in various experimental settings. Meanwhile, we also empirically verify the performance of combining the static and dynamic attentions on open domain multi turn dialogue generation.

Yixuan Wang, Yijun Liu, Shiyu ji et al. · tsinghua

Large language models (LLMs) suffer from high inference latency due to the auto-regressive decoding process. Speculative decoding accelerates inference by generating multiple draft tokens using a lightweight model and verifying them in parallel. However, existing verification methods rely heavily on distributional consistency while overlooking semantic correctness, thereby limiting the potential speedup of speculative decoding. While some methods employ additional models for relaxed verification of draft tokens, they often fail to generalize effectively to more diverse or open-domain settings. In this work, we propose Reflective Verification, a training-free and semantics-aware approach that achieves a better trade-off between correctness and efficiency. Specifically, we leverage the inherent reflective capacity of LLMs to semantically assess the correctness of draft tokens in parallel during verification. Using prompt-based probing, we obtain both the original and reflective distributions of draft tokens in a single forward pass. The fusion of these distributions enables semantic-level verification of draft tokens that incorporates both consistency and correctness. Experiments across multiple domain benchmarks and model scales demonstrate that our method significantly increases the acceptance length of draft tokens without compromising model performance. Furthermore, we find that the proposed Reflective Verification is orthogonal to existing statistical verification methods, and their combination yields additional 5$\sim$15\% improvements in decoding speed.

Qingfu Zhu, Weinan Zhang, Ting Liu et al.

Open-domain dialogue generation suffers from the data insufficiency problem due to the vast size of potential responses. In this paper, we propose to explore potential responses by counterfactual reasoning. Given an observed response, the counterfactual reasoning model automatically infers the outcome of an alternative policy that could have been taken. The resulting counterfactual response synthesized in hindsight is of higher quality than the response synthesized from scratch. Training on the counterfactual responses under the adversarial learning framework helps to explore the high-reward area of the potential response space. An empirical study on the DailyDialog dataset shows that our approach significantly outperforms the HRED model as well as the conventional adversarial learning approaches.

Qingfu Zhu, Lei Cui, Weinan Zhang et al.

Dialogue systems are usually built on either generation-based or retrieval-based approaches, yet they do not benefit from the advantages of different models. In this paper, we propose a Retrieval-Enhanced Adversarial Training (REAT) method for neural response generation. Distinct from existing approaches, the REAT method leverages an encoder-decoder framework in terms of an adversarial training paradigm, while taking advantage of N-best response candidates from a retrieval-based system to construct the discriminator. An empirical study on a large scale public available benchmark dataset shows that the REAT method significantly outperforms the vanilla Seq2Seq model as well as the conventional adversarial training approach.

Weinan Zhang, Ting Liu, Yifa Wang et al.

In this paper, we focus on the personalized response generation for conversational systems. Based on the sequence to sequence learning, especially the encoder-decoder framework, we propose a two-phase approach, namely initialization then adaptation, to model the responding style of human and then generate personalized responses. For evaluation, we propose a novel human aided method to evaluate the performance of the personalized response generation models by online real-time conversation and offline human judgement. Moreover, the lexical divergence of the responses generated by the 5 personalized models indicates that the proposed two-phase approach achieves good results on modeling the responding style of human and generating personalized responses for the conversational systems.

Qingfu Zhu, Weinan Zhang, Lianqiang Zhou et al.

The sequence to sequence architecture is widely used in the response generation and neural machine translation to model the potential relationship between two sentences. It typically consists of two parts: an encoder that reads from the source sentence and a decoder that generates the target sentence word by word according to the encoder's output and the last generated word. However, it faces to the cold start problem when generating the first word as there is no previous word to refer. Existing work mainly use a special start symbol </s>to generate the first word. An obvious drawback of these work is that there is not a learnable relationship between words and the start symbol. Furthermore, it may lead to the error accumulation for decoding when the first word is incorrectly generated. In this paper, we proposed a novel approach to learning to generate the first word in the sequence to sequence architecture rather than using the start symbol. Experimental results on the task of response generation of short text conversation show that the proposed approach outperforms the state-of-the-art approach in both of the automatic and manual evaluations.