L1 P29 · popcorn

compile_fail 29_logsumexp_across_virtual_ranks.py

model

Llama-4-Maverick-17B-128E-Instruct-FP8

turns / max

1 / 1

tool calls

tokens (total)

9084

wall clock

30.9s

truncated

backend

cuda fp32

variant

popcorn

runtime

—

ref runtime

-1.00 μs

speedup

—

started

2026-05-11T08:39:44Z

compile error: Syntax error in custom generated code or ModelNew not found

timingkernelbench.timing

kernel runtime

—

ref runtime

—

speedup

—

kernel mean / median / std

—

ref mean / median / std

—

reference PyTorch source (the prompt the agent sees)

"""
Stable log-sum-exp over virtual ranks: out = logsumexp(x, dim=0), shape (B, D).
"""

import torch
import torch.nn as nn

from kernelbench.distributed_collectives import default_device

R = 4
B = 10
D = 32


class Model(nn.Module):
    def __init__(self):
        super().__init__()

    def forward(self, x: torch.Tensor) -> torch.Tensor:
        return torch.logsumexp(x, dim=0)


def get_inputs():
    dev = default_device()
    p = popcorn_pri
    return [torch.randn(p.jitter_int(R), p.jitter_int(B), p.jitter_int(D, align=8), device=dev)]


def get_init_inputs():
    return []

turn 030.9s turn · 30.9s LLM · 0.0s tools · tokens in=941 · out=8,143 · total=9,084 · FINAL

call → submit_kernel

callsubmit_kernel { "kernel_code": "type=string, description=Full Python source of the ModelNew kernel file. Must be a complete, valid Python module \u2014 not raw CUDA C/C++." }

result ← submit_kernel (fail)

failsubmit_kernel submit_kernel FAILED: kernel did not compile. Syntax error in custom generated code or ModelNew not found

final submitted kernel (1 lines, 134 chars) — full source, untruncated

type=string, description=Full Python source of the ModelNew kernel file. Must be a complete, valid Python module — not raw CUDA C/C++.