RepoBaidu (ERNIE)Baidu (ERNIE)published Jul 31, 2025seen 5d

PaddlePaddle/PassNet

Open original ↗

Captured source

source ↗
GH
GitHub/github.com/PaddlePaddle/PassNet
PaddlePaddle/PassNet repository metadata
Source ↗
published Jul 31, 2025seen 5dcaptured 10hhttp 200method plain

PaddlePaddle/PassNet

Stars: 13

Forks: 15

Open issues: 4

Created: 2025-07-31T12:10:28Z

Pushed: 2026-06-01T02:14:08Z

Default branch: develop

Fork: no

Archived: no

README:

PassNet

PassNet is an AI system for compiler optimization that leverages LLM-driven agents to automatically generate high-performance GPU kernels through compiler pass mechanisms for computation graph optimization. PassNet includes a complete optimization toolchain, the PassBench evaluation benchmark, and the PassAgent agent evaluation framework.

English | [中文](README_cn.md)

Links

Table of Contents

  • [Project Structure](#project-structure)
  • [Architecture Overview](#architecture-overview)
  • [Core Components](#core-components)
  • [DataSet](#dataset)
  • [Quick Start](#quick-start)
  • [PassBench Evaluation Pipeline](#passbench-evaluation-pipeline)
  • [PassAgent Evaluation](#passagent-evaluation)
  • [License](#license)

Project Structure

PassNet/
├── pass_bench/ # PassBench compiler evaluation framework: kernel compilation, correctness verification, performance benchmarking
├── pass_agent/ # PassAgent evaluation framework
├── samples/ # PassBench sample data
├── sample_lists/ # PassBench sample list files (eval/train splits)
├── entry_scripts/ # Evaluation entry scripts
├── graphs/ # Subgraph data
├── graph_lists/ # Subgraph lists and grouping info
├── test/ # Unit tests
├── Dockerfile.nvidia # Docker image definition
└── requirements.txt # Python dependencies

Architecture Overview

┌─────────────────────────────────────────────────────────────────────────┐
│ PassAgent │
│ (LLM-driven Pass Generation) │
│ ┌─────────────────────────────────────────────────────────────────────┐ │◄───┐
│ │ Multi-step Iterative Solving · k-attempts · R2E-Gym Framework │ │ │
│ └─────────────────────────────────────────────────────────────────────┘ │ │
└────────────────┬───────────────────────────────────────┬────────────────┘ │
read data │ generated pass │ │
▼ ▼ │
┌───────────────────────────────────┐ ┌───────────────────────────────┐ │
│ DataSet │ │ PassBench │ │
│ ┌─────────────────────────────┐ │ │ ┌──────────────────────────┐ │ │
│ │ graphs/ │ │ │ │ 1. Execution & Eval │ │ │
│ │ sole_op (5,939) │ │ │ │ Eager Execution │ │ │
│ │ fusible (22,870) │ │ │ │ pass_mgr Execution │ │ │
│ │ typical (25,151) │ │ │ └────────────┬─────────────┘ │ │
│ └─────────────────────────────┘ │ │ │ │ │
│ ┌─────────────────────────────┐ │ │ ▼ │ feedback
│ │ samples/ │ │ │ ┌──────────────────────────┐ │ │
│ │ sole_op (1,029) │ │ │ │ 2. Result Checking │ │ │
│ │ fusible (4,676) │ │ │ │ Correctness & Speedup │ │ │
│ │ typical (4,278) │ │ │ └────────────┬─────────────┘ │ │
│ └─────────────────────────────┘ │ │ │ │ │
│ ┌─────────────────────────────┐ │ │ ▼ │ │
│ │ sample_lists/ │ │ │ ┌──────────────────────────┐ │ │
│ │ train/ │ │ │ │ 3. Score Aggregation │ │ │
│ │ eval/ │ │ │ │ ES(t) & AS Met │ │ │
│ └─────────────────────────────┘ │ │ └──────────────────────────┘ │ │
└───────────────────────────────────┘ └───────────────────────────────┘ │
└─────────────────────┘

Core Components

[PassBench](pass_bench/) — Compiler Evaluation Framework

Provides kernel compilation, correctness verification, and performance benchmarking. It serves as both a standalone evaluation tool and the backend evaluation framework invoked by PassAgent:

  • Kernel Compilation: Executes pass matching and replacement via the pass_mgr compiler method
  • Correctness Verification: Validates numerical correctness of optimized kernels against dtype-specific tolerance thresholds (float32 / float16 / bfloat16)
  • Performance Benchmarking: Measures speedup over 100 trials and outputs aggregated_score.json
  • Score Aggregation: aggregate_es_scores.py computes ES(t) scores across all graphs in a sample

[PassAgent](pass_agent/) — R2E-Gym Agent Evaluation Framework

Evaluates agent capabilities for compiler optimization using the R2E-Gym framework. See [pass_agent/README.md](pass_agent/README.md) for details.

DataSet

graphs — Raw Subgraph Data

Stores raw computation subgraphs extracted from deep learning models, serving as the source for PassBench samples:

  • fusible_subgraphs/: A small set of example fusible subgraphs (1,456), containing computation graphs with multi-operator fusion opportunities
  • hf_subgraphs/ (Legacy): Previous version subgraph data, containing sole op (1,410), fusible (4,167), and typical (6,157) categories
  • hf_subgraphs_v2/: HuggingFace model subgraphs, organized into three categories:
  • sole_op_subgraphs: Single-operator subgraphs (5,939)
  • fusible_subgraphs: Fusible subgraphs (22,870)
  • typical_subgraphs: Typical subgraphs (25,151)

graph_lists — Subgraph Lists and Grouping

Stores subgraph path lists, UID groupings, and other information for sample filtering and group management:

Subgraph Path Lists (line format: subgraph_UID\tsubgraph_relative_path)

| File | Subgraphs | Description | |------|-----------|-------------| | [fusible_subgraphs.txt](graph_lists/fusible_subgraphs.txt) | 1,455 | Example fusible subgraph paths | | [hf_sole_op_subgraphs.txt](graph_lists/hf_sole_op_subgraphs.txt) | 1,410 | Legacy sole op subgraph paths | | [hf_fusible_subgraphs.txt](graph_lists/hf_fusible_subgraphs.txt) | 4,166 | Legacy fusible subgraph paths | | [hf_typical_subgraphs.txt](graph_lists/hf_typical_subgraphs.txt) | 6,157 | Legacy typical subgraph paths | | [hf_sole_op_subgraphs_v2.txt](graph_lists/hf_sole_op_subgraphs_v2.txt) | 5,939 | v2 sole op subgraph paths | | [hf_fusible_subgraphs_v2.txt](graph_lists/hf_fusible_subgraphs_v2.txt) | 22,870 | v2 fusible subgraph paths | | [hf_typical_subgraphs_v2.txt](graph_lists/hf_typical_subgraphs_v2.txt) | 25,151 | v2 typical subgraph paths |

samples — PassBench Evaluation Samples

Evaluation samples generated from graphs/, each serving as an independently executable evaluation unit:

  • fusible_subgraphs/: A small set of example samples from TIMM models' fusible subgraphs, organized by model_name/subgraph_index
  • hf_subgraphs/ (Legacy): Previous version subgraph samples, containing sole op (590), fusible (2,489), and typical (3,382) categories

-…

Excerpt shown — open the source for the full document.

Notability

notability 1.0/10

Low stars, routine repo