Hugging Face Trending Papers

Concordia: JIT-Compiled Persistent-Kernel Checkpointing for Fault-Tolerant LLM Inference

Read the original on Hugging Face Trending Papers →

Long-running LLM agents keep valuable state resident on GPUs: KV caches, request schedulers, communication state, and sometimes online adapters. Losing this state after a GPU or communicator failure can discard minutes to hours of work, yet existing recovery mechanisms either restart the whole serving stack or require application-specific checkpoint logic inside every attention and runtime component.

Summary generated by The Flow from the publisher's feed. The full article lives at Hugging Face Trending Papers.

Hugging Face Trending Papers
Jul 2

DeadPool: Resilient LLM Training with Hot-Swapping via Zero-Overhead Checkpoint

State-of-the-art large language model (LLM) training takes tens of thousands of graphics processing units (GPUs) for months and encounters failures across the software and hardware stack. Existing fault-tolerance mechanisms either impose non-trivial overhead during failure-free execution or suffer from prolonged recovery latency, particularly under scenarios where a small subset of compute nodes experience permanent failures.