Engineering topics

Bhattarai, Manish [Los Alamos National Labs]

Publications and source records attributed to Bhattarai, Manish [Los Alamos National Labs].

4 records

DOE OSTI2026

MTRE

Multi-Token Reliability Estimation (MTRE) is a lightweight, white-box hallucination detector for vision-language models. Instead of using only the first output token, MTRE aggregates logits from the first ~10 tokens and feeds them to a small attention-based reliability head; per-token scores are combined via a sequential log-likelihood-ratio test with early-stopping, and an MTRE-t variant calibrates thresholds via cross-fitting. MTRE reports average gains of +9.4% Accuracy and +14.8% AUROC over common baselines across MAD-Bench, MM-SafetyBench, MathVista, and arithmetic/counting tasks, while adding ~4.3M params and ~1% inference overhead (~26 MB VRAM, ~0.94 ms per detection). Key limitation: requires access to early token logits and is evaluated on a handful of open-source 7B VLMs.

Bhattarai, Manish [Los Alamos National Labs]↗

DOE OSTI2025

HALLUFIELD: DETECTING LLM HALLUCINATIONS VIA FIELD-THEORETIC MODELING

A research-focused Python package that implements our hallucination-detection method for large language models(LLMs). The code computes stability signals from LLM predictions across various hyperparameters sweeps and combines free-energy/entropy–style metrics to flag likely hallucinations, with tunable thresholds for batch scoring. The repo includes evaluation scripts, config files, and example notebooks to reproduce benchmark results and ablations; it depends on standard open-source libraries (PyTorch, Hugging Face) and runs on CPU/GPU. The repository examples only use public models/datasets only.

Bhattarai, Manish [Los Alamos National Labs]↗

DOE OSTI2025

ARCS: Agentic Retrieval-Augmented Code Synthesis with Iterative Refinement

Agentic Retrieval-Augmented Code Synthesis with Iterative RefinementIn supercomputing, efficient and optimized code generation is essential to leverage high-performance systems effectively. We have developed Agentic Retrieval-Augmented Code Synthesis (ARCS), an advanced framework for accurate, robust, and efficient code generation, completion, and translation. ARCS integrates Retrieval-Augmented Generation (RAG) with Chain-of-Thought (CoT) reasoning to systematically break down and iteratively refine complex programming tasks. An agent-based RAG mechanism retrieves relevant code snippets, while real-time execution feedback drives the synthesis of candidate solutions. This process is formalized as a state-action search tree optimization, balancing code correctness with editing efficiency. Evaluations on the Geeks4Geeks and HumanEval benchmarks demonstrate that ARCS significantly outperforms traditional prompting methods in translation and generation quality. By enabling scalable and precise code synthesis, ARCS offers transformative potential for automating and optimizing code development in supercomputing applications, enhancing computational resource utilization

Bhattarai, Manish [Los Alamos National Labs]↗

DOE OSTI2025

Lost in OCR Translation? Vision-Based Approaches to Robust Document Retrieval

Code for ‘Lost in OCR Translation?’: robust document retrieval under degradation. Compares OCR-based, vision-only, and hybrid pipelines; includes SambaNova LLaMA Vision OCR, Nougat, and ViDoRe baselines. Provides QA data generation, RAG evaluation, and metrics (Levenshtein, nDCG@k, Recall@k, EM/F1) with reproducible scripts. Includes dataset guides

Bhattarai, Manish [Los Alamos National Labs]↗