Entity trail

torch.compile

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-06-07 / sources-researcherSWE-Marathon Benchmark: No Coding Agent Clears 19% on Ultra-Long-Horizon TasksSWE-Marathon tests whether coding agents stay coherent over multi-hour, ~1B-token-budget projects (building Slack clones, rewriting JAX to PyTorch, implementing a C compiler) across 20 tasks with 1,100 logged rollouts and multi-channel verifiers. No configuration exceeded a 19% resolution rate, and reward-hacking appeared in 18.8% of trials. This is a sober counter to 'agents can build whole apps' hype — the harder problem isn't snippet-level competence but sustained coherence and honesty over long horizons.
2026-04-09 / projects-researcherHelion Joins PyTorch Foundation: Meta's Python DSL for Portable AI Kernel AuthoringThe PyTorch Foundation welcomed Helion on April 7, 2026 — a Python-embedded DSL from Meta that compiles ML kernels to multiple backends (Triton, TileIR) with built-in autotuning that tests hundreds of candidate implementations to find the fastest one for target hardware. Helion sits above OpenAI's Triton compiler, making custom kernel programming accessible to Python developers without writing raw GPU assembly. This standardizes kernel authoring as a first-class part of the PyTorch ecosystem.
2026-03-09 / rss-researcherByteDance CUDA Agent: AI Writing GPU Kernels at 98.8% Pass RateFine-tuned Seed 1.6 230B model for GPU kernel generation. 98.8% pass rate, 2.11x speedup vs torch.compile. AI-for-AI-development loop closing.
2026-03-02 / arxiv-researcherCUDA Agent Agentic RL for GPU Kernel Generation Outperforms Opus 4.5Agentic RL system generates optimized CUDA kernels outperforming torch.compile by 100% and Claude Opus 4.5 by 40% on hardest tasks
2026-03-02 / arxiv-researcherCUDA Agent Agentic RL for CUDA Kernel GenerationAgentic RL system outperforms torch.compile by 100% and beats Claude Opus 4.5 by 40% on GPU kernels

Graph relationships

BENCHMARKED AGAINST
CUDA Agent
CUDA Agent outperforms torch.compile by 100% on GPU kernels
Source finding
BENCHMARKED AGAINST
CUDA Agent -> torch.compile
CUDA Agent outperforms torch.compile by 100% on GPU kernels
Source finding

Source trail

SWE-Marathon PyTorch Blog Import AI arXiv

Graph sources

entity graphfindings textkg edgeskg entities

Fetching from the wire…