lab NVIDIA Research · just now · research

Relight enables automatic distributed checkpointing and fast-forward replay for task-based systems

from English

Researchers present Relight, a framework that provides automatic, distributed checkpointing for task-based programs along with an efficient fast-forward replay mechanism for full job recovery. Unlike traditional approaches that require manual data migration, Relight leverages the inherent structure of task-based computations to capture and resume state with minimal code changes.

Delivers checkpointing performance and scalability comparable to original, unmodified codes.
Supports operation on up to 512 nodes of the Piz Daint supercomputer.
Eliminates the need for users to manually manage data migration during checkpoint and resume operations.

This approach mitigates risks associated with nondeterministic bugs, hardware faults, and job wall-time limits in long-running programs by automating state management.

Importance 1/3 Trust 3/3 NVIDIA Research Research paper

Read original

arxiv arXiv cs.CL · 1d ago

STAR metric and StarPO framework improve document-to-document machine translation alignment

Researchers introduce Sentence Translation Alignment Rate (STAR), an auxiliary metric designed to quantify sentence-level structural fidelity in document-to-document machine translation, addressing issues like omissions and hallucinations. They propose STAR-masked Preference Optimization (StarPO), a framework that ranks hypotheses by structural quality and uses a dynamic alignment mask to focus optimization on misaligned segments.

media Hugging Face Forums · 1d ago · 2 views

Ontological inversion: cross-model vector memory for steering and recall

A preprint by independent researcher Jason Van Pham proposes translating continuous representations between frozen models to enable steering, recall, and reasoning without weight editing. The work introduces a method where concepts are compiled into residual directions in one model's space and mapped via an affine transformation into another model's hidden layers.

arxiv arXiv cs.CL · 2d ago · 5 views

JIT-Agent enables on-the-fly harness evolution to boost agentic LLM performance

The authors present JIT-Agent, a harness intelligence model trained to synthesize task-adaptive agent harnesses for arbitrary off-the-shelf agentic LLMs. This approach formalizes the agent harness as a composable artifact governed by a fixed four-module protocol, allowing the system to customize and repair harnesses on the fly.

media Hugging Face Forums · 2d ago · 3 views

Gemma 4 baseline trace shows hallucination loop and self-report

A baseline trace of Gemma 4 with zero force cap reveals a persistent hallucination loop where the model generates a self-report of failure before entering a repetition well. The telemetry confirms that role-tag bleed and phrases like "STOPITSTOP" occur within this unstable state, mirroring patterns observed in Llama and Gemma 3 loaders since December 2025.

media Hugging Face Forums · 2d ago · 3 views

Study finds progressive representational structuring across depth in 8 open models

A new preprint analyzes hidden-state dynamics in eight locally instrumented small open Transformer models to determine if internal representations follow a structured progression during inference. The study distinguishes between changes occurring across model depth and those evolving over autoregressive generation time.