Now in private alpha

From arXiv link to running code in minutes

PaperRun turns machine learning papers into runnable, reproducible repositories. Paste a link, and our agent resolves the environment, wires up the reference implementation, and verifies the paper's core claims.

Free for students and academic labs during alpha.

Thanks. You are on the list, we will reach out soon.

$ paperrun reproduce arxiv.org/abs/2403.00835
 Parsing paper........................ done
 Resolving environment (py3.11, torch 2.3) done
 Locating reference implementation.... done
 Building repro repo.................. done
 Running headline experiment (Table 2)  3.1x speedup, matches claim

✓ Repo ready: ./cllm-repro  (runs in 1 command)

Reproduction without the archaeology

Most ML papers take days to get running: broken dependencies, missing configs, undocumented preprocessing. PaperRun automates that entire slog.

Paste a paper

Drop in an arXiv link or PDF. PaperRun extracts the method, hyperparameters, datasets, and evaluation setup.

Agent builds the repo

An agent resolves a working environment, pulls or reconstructs the implementation, and pins every dependency.

Run and verify

One command runs the headline experiment and reports how the results compare against the paper's claims.

Built for researchers, by researchers

Started because we were tired of losing whole weekends to "pip install -r requirements.txt" failing on line one.

⚙︎

Environment resolution

Automatic detection of framework versions, CUDA constraints, and dataset dependencies, pinned into a reproducible lockfile.

Claim verification

Headline experiments re-run automatically, with a side-by-side report of reproduced metrics versus published numbers.

Runs anywhere

Export a container or a plain repo. Run locally, on your lab cluster, or on rented GPUs with no changes.

Annotated code

Every module linked back to the equation or section of the paper it implements, so reading the code teaches you the paper.

Baselines on tap

Comparing against prior work? Pull reproducible baselines from papers we have already built, instead of rebuilding them yourself.

Lab workspaces

Shared reproduction library for your research group, so nobody rebuilds an environment a labmate already fixed.

Stop reproducing. Start researching.

Join the waitlist for the private alpha. Free for students and academic labs.

Thanks. You are on the list, we will reach out soon.