BFCL
PulseAugur coverage of BFCL — every cluster mentioning BFCL across labs, papers, and developer communities, ranked by signal.
6 day(s) with sentiment data
-
New DM-Call framework improves LLM function calling accuracy
Researchers have introduced DM-Call, a novel framework designed to enhance the reliability of function calling in large-language models (LLMs). This system integrates local schema validation with a separate decision-mod…
-
LIBRA framework enhances LLM agent tool retrieval via programmatic library induction
Researchers have introduced LIBRA, a framework designed to programmatically create executable tool libraries from flat catalogs for large language model agents. LIBRA uses an LLM to propose tool contracts, which are the…
-
New CGAP method enhances AI agent planning and tool execution
Researchers have developed a new method called Capability-Grounded Agentic Planning (CGAP) to improve how AI agents execute tasks using tools. Unlike traditional trial-and-error approaches, CGAP deterministically compil…
-
New DPDM architecture boosts agent decision accuracy using frozen predictors
Researchers have introduced Decoupled Predictive Decision Making (DPDM), a novel architecture designed to enhance agent decision-making accuracy. DPDM separates semantic representation, handled by a language model like …
-
AI agents struggle with subtle tool errors, study finds
A new study titled "Loud Failures, Quiet Failures" by Obada Kraishan investigates how AI agents handle errors when interacting with tools. The research found that agents are significantly better at detecting explicit er…
-
Underdog Saluki 27B: A guide to running a quantized Qwen3.8-27B model
This tutorial details how to run the Underdog Saluki 27B model, a quantized version of Qwen3.8-27B. Saluki is noted for maintaining tool-calling capabilities at a reduced size and reportedly outperforms the full model o…
-
Trajectory2Gym framework enhances tool-use agent training with decision grounding · 2 sources tracked
Researchers have developed Trajectory2Gym, a novel framework designed to improve the training of tool-use agents. This system focuses on constructing interactions that ensure compositional coverage and decision groundin…
-
New BRACE method optimizes language model training with reward disagreement
Researchers have developed a new method called BRACE (Bank-calibrated Rollout Allocation under Correlated Episodes) to optimize the training of tool-using language models, specifically within Model Context Protocol (MCP…
-
New PORA method optimizes LLM agent reasoning for efficiency
Researchers have developed a new method called Paired Outcome-guided Reasoning Allocation (PORA) to help AI agents decide when to use more reasoning power. PORA works by training agents to compare the outcomes of both s…
-
Tīrtha Bench launches to evaluate LLM tool-calling for coding agents
A new benchmark called Tīrtha Bench has been released to evaluate Large Language Models (LLMs) specifically for coding agent tasks. Unlike traditional leaderboards, Tīrtha Bench uses a single, consistent test set of 800…
-
Autotrust releases two decision-making AI models evaluated on Decision Index
Autotrust has released two new models, AutoTrust-JEV-27B-v1.1 and AutoTrust-Decider-Pro-Vision-27B-v1.1, which are designed for structured decision-making. These models assign probabilities to answer options and have be…
-
New 'Handoff Curves' method optimizes multi-turn LLM routing
Researchers have developed a new method called "Handoff Curves" to analyze and optimize multi-turn routing in Large Language Models (LLMs). This approach evaluates how initial model selection, subsequent handoffs, and t…
-
New framework enhances small language model agents with selective cloud escalation
Researchers have developed STEPGATE, a framework designed to improve the efficiency of small language model (SLM) agents. This system intelligently assesses the difficulty of each step in an agent's task and selectively…
-
Google's Gemma 4 models repacked for enhanced performance on single TPU v5e
A technical guide details how to repack Google's quantization-aware-trained (QAT) Gemma 4 models for improved performance on a single Google Cloud TPU v5e chip. The repacked models, particularly the 12B parameter versio…
-
Imprint Reader model deciphers language model learning from weight updates
Researchers have developed the Imprint Reader, a model designed to interpret the learning process of other language models by analyzing their weight updates. This model, trained using Semantic Mount-and-Read Tuning (SaR…
-
New benchmark and data-synthesis method improve LLM multi-step tool-calling
Researchers have introduced KOPA-Bench, a new benchmark designed to evaluate the performance of open-source LLM agents in executing multi-step tool-calls over Korean public APIs. To address the current underperformance …
-
AI research advances inference, optimization, and mobile benchmarking
Researchers are exploring advanced techniques for improving AI inference and statistical analysis, particularly in resource-constrained environments. One paper introduces IMABO, a framework for Online Hyperparameter Opt…
-
AI agent benchmarks audited for noise, SIGMA framework tackles multi-agent robustness
Two new research papers explore challenges in AI agent performance and robustness. The first paper introduces SIGMA, a hierarchical framework designed to improve multi-agent reinforcement learning by accounting for stru…
-
Nanbeige4.2-3B model fixed for Apple Silicon deployment
A new paper details the challenges and solutions for deploying the Nanbeige4.2-3B model, a 3-billion parameter agentic model utilizing a Looped Transformer architecture, on Apple Silicon. Researchers identified five cri…
-
New research explores modular and recursive methods for automatic prompt optimization
Two new research papers introduce novel methods for optimizing prompts used with large language models. The first, SAPO, breaks down prompts into segments like role, context, and task, allowing for targeted improvements…