Kinjal Basu

Kinjal Basu, Ibrahim Abdelaziz, Kiran Kate et al. · ibm-research

The resurgence of autonomous agents built using large language models (LLMs) to solve complex real-world tasks has brought increased focus on LLMs' fundamental ability of tool or function calling. At the core of these agents, an LLM must plan, execute, and respond using external tools, APIs, and custom functions. Research on tool calling has gathered momentum, but evaluation benchmarks and datasets representing the complexity of the tasks have lagged behind. In this work, we focus on one such complexity, nested sequencing, with the goal of extending existing benchmarks and evaluation. Specifically, we present NESTFUL, a benchmark to evaluate LLMs on nested sequences of API calls, i.e., sequences where the output of one API call is passed as input to a subsequent call. NESTFUL contains 1800+ nested sequences where all the function calls are executable. Experimental results on a variety of models show that the best-performing model (GPT-4o) achieves a full sequence match accuracy of 28% and a win-rate of 60%, necessitating a large scope for improvement in the nested sequencing aspect of function calling. Our analysis of these results provides possible future research directions for the community, in addition to a benchmark to track progress. We have released the NESTFUL dataset under the Apache 2.0 license at https://github.com/IBM/NESTFUL.

Maxwell Crouse, Ibrahim Abdelaziz, Ramon Astudillo et al. · ibm-research

Autonomous, goal-driven agents powered by LLMs have recently emerged as promising tools for solving challenging problems without the need for task-specific finetuned models that can be expensive to procure. Currently, the design and implementation of such agents is ad hoc, as the wide variety of tasks that LLM-based agents may be applied to naturally means there can be no one-size-fits-all approach to agent design. In this work we aim to alleviate the difficulty of designing and implementing new agents by proposing a minimalistic generation framework that simplifies the process of building agents. The framework we introduce allows the user to define desired agent behaviors in a high-level, declarative specification that is then used to construct a decoding monitor which guarantees the LLM will produce an output exhibiting the desired behavior. Our declarative approach, in which the behavior is described without concern for how it should be implemented or enforced, enables rapid design, implementation, and experimentation with different LLM-based agents. We demonstrate how the proposed framework can be used to implement recent LLM-based agents (e.g., ReACT), and show how the flexibility of our approach can be leveraged to define a new agent with more complex behavior, the Plan-Act-Summarize-Solve (PASS) agent. Lastly, we demonstrate that our method outperforms other agents on multiple popular reasoning-centric question-answering benchmarks.

Mayank Mishra, Matt Stallone, Gaoyuan Zhang et al. · ibm-research

Large Language Models (LLMs) trained on code are revolutionizing the software development process. Increasingly, code LLMs are being integrated into software development environments to improve the productivity of human programmers, and LLM-based agents are beginning to show promise for handling complex tasks autonomously. Realizing the full potential of code LLMs requires a wide range of capabilities, including code generation, fixing bugs, explaining and documenting code, maintaining repositories, and more. In this work, we introduce the Granite series of decoder-only code models for code generative tasks, trained with code written in 116 programming languages. The Granite Code models family consists of models ranging in size from 3 to 34 billion parameters, suitable for applications ranging from complex application modernization tasks to on-device memory-constrained use cases. Evaluation on a comprehensive set of tasks demonstrates that Granite Code models consistently reaches state-of-the-art performance among available open-source code LLMs. The Granite Code model family was optimized for enterprise software development workflows and performs well across a range of coding tasks (e.g. code generation, fixing and explanation), making it a versatile all around code model. We release all our Granite Code models under an Apache 2.0 license for both research and commercial use.

Yankai Zeng, Abhiramon Rajashekharan, Kinjal Basu et al.

The development of large language models (LLMs), such as GPT, has enabled the construction of several socialbots, like ChatGPT, that are receiving a lot of attention for their ability to simulate a human conversation. However, the conversation is not guided by a goal and is hard to control. In addition, because LLMs rely more on pattern recognition than deductive reasoning, they can give confusing answers and have difficulty integrating multiple topics into a cohesive response. These limitations often lead the LLM to deviate from the main topic to keep the conversation interesting. We propose AutoCompanion, a socialbot that uses an LLM model to translate natural language into predicates (and vice versa) and employs commonsense reasoning based on Answer Set Programming (ASP) to hold a social conversation with a human. In particular, we rely on s(CASP), a goal-directed implementation of ASP as the backend. This paper presents the framework design and how an LLM is used to parse user messages and generate a response from the s(CASP) engine output. To validate our proposal, we describe (real) conversations in which the chatbot's goal is to keep the user entertained by talking about movies and books, and s(CASP) ensures (i) correctness of answers, (ii) coherence (and precision) during the conversation, which it dynamically regulates to achieve its specific purpose, and (iii) no deviation from the main topic.

Kinjal Basu, Ibrahim Abdelaziz, Subhajit Chaudhury et al. · ibm-research

There is a growing need for Large Language Models (LLMs) to effectively use tools and external Application Programming Interfaces (APIs) to plan and complete tasks. As such, there is tremendous interest in methods that can acquire sufficient quantities of train and test data that involve calls to tools / APIs. Two lines of research have emerged as the predominant strategies for addressing this challenge. The first has focused on synthetic data generation techniques, while the second has involved curating task-adjacent datasets which can be transformed into API / Tool-based tasks. In this paper, we focus on the task of identifying, curating, and transforming existing datasets and, in turn, introduce API-BLEND, a large corpora for training and systematic testing of tool-augmented LLMs. The datasets mimic real-world scenarios involving API-tasks such as API / tool detection, slot filling, and sequencing of the detected APIs. We demonstrate the utility of the API-BLEND dataset for both training and benchmarking purposes.

Kinjal Basu, Keerthiram Murugesan, Subhajit Chaudhury et al. · ibm-research

Text-based games (TBGs) have emerged as an important collection of NLP tasks, requiring reinforcement learning (RL) agents to combine natural language understanding with reasoning. A key challenge for agents attempting to solve such tasks is to generalize across multiple games and demonstrate good performance on both seen and unseen objects. Purely deep-RL-based approaches may perform well on seen objects; however, they fail to showcase the same performance on unseen objects. Commonsense-infused deep-RL agents may work better on unseen data; unfortunately, their policies are often not interpretable or easily transferable. To tackle these issues, in this paper, we present EXPLORER which is an exploration-guided reasoning agent for textual reinforcement learning. EXPLORER is neurosymbolic in nature, as it relies on a neural module for exploration and a symbolic module for exploitation. It can also learn generalized symbolic policies and perform well over unseen data. Our experiments show that EXPLORER outperforms the baseline agents on Text-World cooking (TW-Cooking) and Text-World Commonsense (TWC) games.

Ibrahim Abdelaziz, Kinjal Basu, Mayank Agarwal et al.

Large language models (LLMs) have recently shown tremendous promise in serving as the backbone to agentic systems, as demonstrated by their performance in multi-faceted, challenging benchmarks like SWE-Bench and Agent-Bench. However, to realize the true potential of LLMs as autonomous agents, they must learn to identify, call, and interact with external tools and application program interfaces (APIs) to complete complex tasks. These tasks together are termed function calling. Endowing LLMs with function calling abilities leads to a myriad of advantages, such as access to current and domain-specific information in databases and knowledge sources, and the ability to outsource tasks that can be reliably performed by tools, e.g., a Python interpreter or calculator. While there has been significant progress in function calling with LLMs, there is still a dearth of open models that perform on par with proprietary LLMs like GPT, Claude, and Gemini. Therefore, in this work, we introduce the GRANITE-20B-FUNCTIONCALLING model under an Apache 2.0 license. The model is trained using a multi-task training approach on seven fundamental tasks encompassed in function calling, those being Nested Function Calling, Function Chaining, Parallel Functions, Function Name Detection, Parameter-Value Pair Detection, Next-Best Function, and Response Generation. We present a comprehensive evaluation on multiple out-of-domain datasets comparing GRANITE-20B-FUNCTIONCALLING to more than 15 other best proprietary and open models. GRANITE-20B-FUNCTIONCALLING provides the best performance among all open models on the Berkeley Function Calling Leaderboard and fourth overall. As a result of the diverse tasks and datasets used for training our model, we show that GRANITE-20B-FUNCTIONCALLING has better generalizability on multiple tasks in seven different evaluation datasets.

Suraj Kothawade, Vinaya Khandelwal, Kinjal Basu et al.

Driving an automobile involves the tasks of observing surroundings, then making a driving decision based on these observations (steer, brake, coast, etc.). In autonomous driving, all these tasks have to be automated. Autonomous driving technology thus far has relied primarily on machine learning techniques. We argue that appropriate technology should be used for the appropriate task. That is, while machine learning technology is good for observing and automatically understanding the surroundings of an automobile, driving decisions are better automated via commonsense reasoning rather than machine learning. In this paper, we discuss (i) how commonsense reasoning can be automated using answer set programming (ASP) and the goal-directed s(CASP) ASP system, and (ii) develop the AUTO-DISCERN system using this technology for automating decision-making in driving. The goal of our research, described in this paper, is to develop an autonomous driving system that works by simulating the mind of a human driver. Since driving decisions are based on human-style reasoning, they are explainable, their ethics can be ensured, and they will always be correct, provided the system modeling and system inputs are correct.

Kinjal Basu, Huaduo Wang, Nancy Dominguez et al.

We report on the design and development of the CASPR system, a socialbot designed to compete in the Amazon Alexa Socialbot Challenge 4. CASPR's distinguishing characteristic is that it will use automated commonsense reasoning to truly "understand" dialogs, allowing it to converse like a human. Three main requirements of a socialbot are that it should be able to "understand" users' utterances, possess a strategy for holding a conversation, and be able to learn new knowledge. We developed techniques such as conversational knowledge template (CKT) to approximate commonsense reasoning needed to hold a conversation on specific topics. We present the philosophy behind CASPR's design as well as details of its implementation. We also report on CASPR's performance as well as discuss lessons learned.

Fang Li, Huaduo Wang, Kinjal Basu et al.

We consider the problem of finding relevant consistent concepts in a conversational AI system, particularly, for realizing a conversational socialbot. Commonsense knowledge about various topics can be represented as an answer set program. However, to advance the conversation, we need to solve the problem of finding relevant consistent concepts, i.e., find consistent knowledge in the "neighborhood" of the current topic being discussed that can be used to advance the conversation. Traditional ASP solvers will generate the whole answer set which is stripped of all the associations between the various atoms (concepts) and thus cannot be used to find relevant consistent concepts. Similarly, goal-directed implementations of ASP will only find concepts directly relevant to a query. We present the DiscASP system that will find the partial consistent model that is relevant to a given topic in a manner similar to how a human will find it. DiscASP is based on a novel graph-based algorithm for finding stable models of an answer set program. We present the DiscASP algorithm, its implementation, and its application to developing a conversational socialbot.

Kinjal Basu, Sarat Chandra Varanasi, Farhad Shakerin et al.

Understanding the meaning of a text is a fundamental challenge of natural language understanding (NLU) and from its early days, it has received significant attention through question answering (QA) tasks. We introduce a general semantics-based framework for natural language QA and also describe the SQuARE system, an application of this framework. The framework is based on the denotational semantics approach widely used in programming language research. In our framework, valuation function maps syntax tree of the text to its commonsense meaning represented using basic knowledge primitives (the semantic algebra) coded using answer set programming (ASP). We illustrate an application of this framework by using VerbNet primitives as our semantic algebra and a novel algorithm based on partial tree matching that generates an answer set program that represents the knowledge in the text. A question posed against that text is converted into an ASP query using the same framework and executed using the s(CASP) goal-directed ASP system. Our approach is based purely on (commonsense) reasoning. SQuARE achieves 100% accuracy on all the five datasets of bAbI QA tasks that we have tested. The significance of our work is that, unlike other machine learning based approaches, ours is based on "understanding" the text and does not require any training. SQuARE can also generate an explanation for an answer while maintaining high accuracy.

Kinjal Basu

Our research is focused on making a human-like question answering system which can answer rationally. The distinguishing characteristic of our approach is that it will use automated common sense reasoning to truly "understand" dialogues, allowing it to converse like a human. Humans often make many assumptions during conversations. We infer facts not told explicitly by using our common sense. Incorporating commonsense knowledge in a question answering system will simply make it more robust.