Binesh Sadanandan Dissertation
Read dissertation

Reproducibility

What it cost, and what you need to run it again

This work took about ~4,800 GPU-hours on NVIDIA A100 80GB hardware. Most of that went on evaluation rather than training, and the headline mitigation is one of the cheapest things in the dissertation to reproduce: about twenty minutes on a single GPU. This page itemizes the cost, pins the software and seeds, and says plainly which artifacts are public and which need a request.

Figures on this page are transcribed from the dissertation's appendices (dissertation/chapters/08_appendix.tex, Appendix B (Computational Resources, Reproducibility) and Appendix C (Software Environment, Hardware, Random Seeds)).

Where the compute went

Total across all thrusts on a shared cluster, including failed and exploratory runs, not only the runs that appear in the dissertation. Every run was tracked in Weights & Biases.

PSF-Med benchmark evaluation and equivalence audit 1,300 h · 27%

6 models, 3 datasets, roughly 93k paraphrase pairs; rubric-based judge audit of 122,778 pairs; regeneration of the PadChest paraphrase set

Uncertainty quantification pipeline 950 h · 20%

5 models, 2 datasets; softmax entropy, Monte Carlo dropout, temperature scaling, deep ensembles across 5 seeds, and text-only baselines

Corruption robustness sweep 600 h · 13%

5 corruption types, 3 severities, across models, datasets, and uncertainty methods

Safety evaluation and deployment audit 525 h · 11%

Multiple model families; image-swap, text-only, and null-image controls

Attention grounding and causal analyses 525 h · 11%

Attention extraction, patch occlusion, region-of-interest ablation, and laterality across models

LoRA training, layer and block sweeps, replication study 450 h · 9%

Includes the five-seed replication study

Mechanistic interpretability 450 h · 9%

Sparse-autoencoder screening and transfer, activation and residual-transplant patching, canonical validation, and lens fitting

Total: ~4,800 GPU-hours (~200 GPU-days). The line items above sum to 4,800 hours; the reported total is rounded and includes runs that produced nothing publishable.

What it costs to reproduce one result

The aggregate above is the cost of the whole programme, including exploration. Reproducing a single headline result is far cheaper, and that asymmetry is deliberate: the interventions in this dissertation are small.

Entry cost for individual results
TaskCostDetail
Reproduce the targeted-LoRA mitigation~20 minutes on one A100~24 GB peak memory, 4.38M trainable parameters (0.10%). This is the headline result and it is cheap to reproduce.
Reproduce the full-LoRA comparison~45 minutes on one A10030.5M trainable parameters (0.72%)
Run sparse-autoencoder extractionsingle A100, ~4 GBSAE inference only; the Gemma Scope weights are downloaded, not trained
Score the deployment gate on one prediction5 to 6 forward passes, ~600-900 msK = 3 paraphrases plus the image-reliance checks, on an A100

Environment, hardware, and seeds

Software

Python
3.10.12
PyTorch
2.6.0 with CUDA 12.4
Transformers
4.57.3
PEFT
0.7.0
scikit-learn
1.3.2
scipy
1.11.4
Sparse autoencoders
custom implementation over Gemma Scope weights

Hardware

  • Training: single NVIDIA A100 80GB, 64 GB system RAM
  • Inference: NVIDIA A100 80GB or NVIDIA A6000 48GB
  • Multi-GPU: 8x A100 for the corruption sweep
  • Sparse-autoencoder extraction: single A100, about 4 GB

Seeds

Primary seed 42; the replication study uses 42, 123, 456, 789, 2024.

The primary seed is 42 for all main results; the replication study uses all five. Model weights are loaded from Hugging Face checkpoints with commit hashes recorded in the experiment configurations.

What is already public

These need no request. The underlying chest X-ray images are never redistributed here: they come from their original providers under their own agreements.

Public artifacts
ArtifactWhat it isLink
PSF-Med benchmark codeBenchmark construction and evaluation code for PSF-Med: paraphrase generation, equivalence auditing, and flip-rate evaluation (Thrust 1).Open
PSF-Med dataset (Hugging Face)The released PSF-Med paraphrase benchmark data on Hugging Face; image files are not redistributed and must be obtained from the original credentialed sources.Open
Mechanistically-guided LoRA weights (Hugging Face collection)Targeted and full LoRA adapter checkpoints for MedGemma-4B from the CHIL 2026 paper, trained at n=12K.Open
Targeted MedGemma-4B LoRA modelThe n=12K targeted LoRA adapter for MedGemma-4B used in the paraphrase-consistency mitigation experiments.Open
Paraphrase-consistency mitigation codeTraining and evaluation code for the targeted LoRA mitigation and its safety re-audit (Thrust 3).Open
VSF-Med project siteProject page for VSF-Med, the vulnerability scoring framework companion work on medical VLM safety.Open
VSF-Med codeImplementation of the VSF-Med vulnerability scoring framework and its evaluation pipeline.Open
MedMCQA robustness study codeCode for the companion study of prompt sensitivity and chain-of-thought prompting on medical question answering.Open
Predictive entropy study codeCode and reproducibility materials for the UNSURE 2026 predictive-entropy screen for errors and paraphrase instability.Open
Attention Without Grounding codeCode and data instructions for the iMIMIC 2026 causal evaluation of visual explanations in medical VLMs.Open

Requesting the rest

Public artifacts need no request; the link is on this page. A request is only needed for the parts that are not yet public, and it exists so that access is recorded and so that the code can be sent with the context needed to run it.

Available on request
ArtifactWhat it isAccess
Weights & Biases experiment trackingExperiment tracking workspace for the uncertainty-quantification and robustness runs.On request
medical-vlm-robustness research repositoryThe full research monorepo behind the dissertation (pipelines, analyses, dissertation sources); private during the pre-defense period.private

Send a request

The button opens a mail draft addressed to the author with the questions below already in it. Nothing is submitted from this page and nothing is logged here: you will see the message before it is sent.

  1. Your name, role, and institution
  2. Which artifact you want (benchmark code, LoRA training, safety pipeline, mechanistic analyses, or a specific figure or table)
  3. What you intend to do with it
  4. Whether you already hold the credentials for the underlying image datasets (PhysioNet for MIMIC-CXR and VinDr-CXR, BIMCV for PadChest)

Open a request draft All papers and resources

Requests go to contact@bineshkumar.me. Access to the chest X-ray images themselves cannot be granted by the author: MIMIC-CXR and VinDr-CXR are credentialed through PhysioNet, and PadChest through BIMCV.

Cite

Permanent link

Type to search. Press Escape to close.