Files

T

coolguyandCodex 847cc84c9f try torch jl mlp criterion retry

Co-Authored-By: Codex <codex@openai.com>

2026-05-07 21:36:21 +09:00

src

try torch jl mlp criterion retry

2026-05-07 21:36:21 +09:00

bench_torch.jl

try torch jl mlp criterion retry

2026-05-07 21:36:21 +09:00

Manifest.toml

try torch jl mlp criterion retry

2026-05-07 21:36:21 +09:00

Project.toml

try torch jl mlp criterion retry

2026-05-07 21:36:21 +09:00

README.md

try torch jl mlp criterion retry

2026-05-07 21:36:21 +09:00

results.json

try torch jl mlp criterion retry

2026-05-07 21:36:21 +09:00

README.md

Julia Torch MLP Forward Benchmark

Attempted Criterion 4 re-measurement using Torch.jl for the same DeepCFRMLP shape used in experiments/julia_flux_mlp/:

input dim: 365
hidden size: 512
hidden layers: 3
output dim: 22
activation: ReLU

Run:

tools/julia/current/bin/julia --project=experiments/julia_torch_mlp \
  experiments/julia_torch_mlp/bench_torch.jl

Result (2026-05-07)

Torch.jl could not be loaded on this host:

Torch.jl v0.1.3 failed to precompile:
UndefVarError: libtorch_c_api not defined in Torch.Wrapper

This happens before any model construction or timing is possible, so no DeepCFRMLP forward benchmark could be run.

Verdict: BLOCKED. Criterion 4 remains FAIL based on the completed Flux.jl measurement. Torch.jl does not reverse the decision because it is not usable in the current Julia 1.11.9 environment.