Large language models (LLMs) often struggle with mathematical problems that require exact computation or multi-step algebraic reasoning. Tool-integrated reasoning (TIR) offers a promising solution by leveraging external tools such as code interpreters to ensure correctness, but it introduces inference-time dependencies that hinder scalability and deployment. In this work, we propose a new paradigm for distilling tool knowledge into LLMs purely through natural language. We first construct a Solver Agent that solves math problems by interleaving planning, symbolic tool calls, and reflective reasoning. Then, using a back-translation pipeline powered by multiple LLM-based agents, we convert interleaved TIR traces into natural language reasoning traces. A Translator Agent generates explanations for individual tool calls, while a Rephrase Agent merges them into a fluent and globally coherent narrative. Empirically, we show that fine-tuning a small open-source model on these synthesized traces enables it to internalize both tool knowledge and structured reasoning patterns, yielding gains on competition-level math benchmarks without requiring tool access at inference.

This study investigates the structured generation capabilities of large language models (LLMs), focusing on producing valid JSON outputs against a given schema. Despite the widespread use of JSON in integrating language models with programs, there is a lack of comprehensive analysis and benchmarking of these capabilities. We explore various aspects of JSON generation, such as structure understanding, escaping, and natural language description, to determine how to assess and enable LLMs to generate valid responses. Building upon this, we propose SchemaBench features around 40K different JSON schemas to obtain and assess models' abilities in generating valid JSON. We find that the latest LLMs are still struggling to generate a valid JSON string. Moreover, we demonstrate that incorporating reinforcement learning with a Fine-grained Schema Validator can further enhance models' understanding of JSON schema, leading to improved performance. Our models demonstrate significant improvement in both generating JSON outputs and downstream tasks.

High-quality, large-scale instructions are crucial for aligning large language models (LLMs), however, there is a severe shortage of instruction in the field of natural language understanding (NLU). Previous works on constructing NLU instructions mainly focus on information extraction (IE), neglecting tasks such as machine reading comprehension, question answering, and text classification. Furthermore, the lack of diversity in the data has led to a decreased generalization ability of trained LLMs in other NLU tasks and a noticeable decline in the fundamental model's general capabilities. To address this issue, we propose Hum, a large-scale, high-quality synthetic instruction corpus for NLU tasks, designed to enhance the NLU capabilities of LLMs. Specifically, Hum includes IE (either close IE or open IE), machine reading comprehension, text classification, and instruction generalist tasks, thereby enriching task diversity. Additionally, we introduce a human-LLMs collaborative mechanism to synthesize instructions, which enriches instruction diversity by incorporating guidelines, preference rules, and format variants. We conduct extensive experiments on 5 NLU tasks and 28 general capability evaluation datasets for LLMs. Experimental results show that Hum enhances the NLU capabilities of six LLMs by an average of 3.1\%, with no significant decline observed in other general capabilities.

Natural language understanding includes the tasks of intent detection (identifying a user's objectives) and slot filling (extracting the entities relevant to those objectives). Prior slot filling methods assume that each intent type cannot occur more than once within a message, however this is often not a valid assumption for real-world settings. In this work, we generalize slot filling by removing the constraint of unique intents in a message. We cast this as a JSON generation task and approach it using a language model. We create a pre-training dataset by combining DBpedia and existing slot filling datasets that we convert for JSON generation. We also generate an in-domain dataset using GPT-3. We train T5 models for this task (with and without exemplars in the prompt) and find that both training datasets improve performance, and that the model is able to generalize to intent types not seen during training.

This plugin contains C, Blueprint and python scripts that encapsulate TensorFlow operations as an Actor Component. It depends on an UnrealEnginePython plugin fork and the SocketIO Client plugin; these are always included in binary releases so no manual external downloading is necessary. See Note on Dependencies section for details on implementation and architecture. See unreal forum thread for discussions. There is currently only a working build for the Windows platform.

It has been always difficult to consume TensorFlow or ONNX models without the help of tools like TensorFlow Serving or gRPC and all the fun that comes with protocol buffers. Hosting deep learning models to be consumed using REST was very hard although this is probably the most common approach application developers would start with. Microsoft has recently released Azure Machine Learning service which comes with heaps of features to facilitate development and deployment of machine learning models. One of those features is hosting ONNX models in docker containers to be consumed using REST. In this post, we go through an end to end workflow of hosting a sample ONNX model and consuming it from a .NET application.

You will be able to build a model to generate 5-star Yelp reviews like those. Training the model could easily take up a couple of days even on GPU. Luckily the pre-trained model weights are available. So we could jump directly to the fun part to generate reviews. The Yelp Dataset is freely available in JSON format.

This plugin source contains C, Blueprint and python scripts that encapsulate TensorFlow operations as an Actor Component. The plugin depends on UnrealEnginePython plugin fork and SocketIO Client plugin. Releases for this plugin contain compiled versions of all dependency plugins and you should be able to drag and drop it into your project. If you have ideas or fixes, consider contributing! See unreal forum thread for discussions.

"OK Google, what can you do when connected to Talend?" In this tutorial, I will show how to create an Agent in API.AI that will respond to commands spoken to Google Home. The Agent will reverse the words in a sentence spoken to Google Home by making use of a Talend web service which is used to carry out the word reversal. A very simple example, but it demonstrates the ground work you will need to create some really quite interesting applications. You do not need one to try this tutorial out as Google has provided an emulator, but I can highly recommend the device. Recently Google opened up access to the Actions on Google API. You can either use the Actions SDK or use API.AI. API.AI was recently acquired by Google. While API.AI is really quite simple to use, it is quite limited in how it can be used with Google Home at the moment.