Txus Bach PRO

txus

duksis's profile picture

seanmanifest's profile picture

quicola's profile picture

https://double-dissent.fika.bar/

txus
txus

AI & ML interests

Transformers, RL

Recent Activity

updated a collection about 17 hours ago

Reproduction: Rare Event Analysis of LLMs (ICML 2026 #71154)

updated a collection about 17 hours ago

Reproduction: Rare Event Analysis of LLMs (ICML 2026 #71154)

updated a Space about 17 hours ago

txus/71154

View all activity

Organizations

txus 's collections 6

Reproduction: Rare Event Analysis of LLMs (ICML 2026 #71154)

Independent reproduction of arXiv 2602.06791 (Dorman et al.). MBAR/TPS rare-event estimation for TinyStories-8M, ARI & Log-Prob observables.

txus/rea-llm-repro-71154

Updated about 17 hours ago
Running

Repro: Rare Event Analysis of Large Language Models

🎯

Collaborate on LLM rare-event logs with a coding agent
Rare Event Analysis of Large Language Models

Paper • 2602.06791 • Published Feb 6

Repro: Janus-LoRA (ICML 2026, XJ5r19iN9G)

Reproduction of Janus-LoRA — all 6 claims. Logbook Space, GPU-run dataset, and artifact bucket.

Running

Repro - Janus-LoRA: A Balanced Low-Rank Adaptation for Continual Learning

⚖

Explore experiment logs and sync them with a coding agent
txus/janus-lora-repro

Updated 2 days ago • 48

Reproduction: L-SR1 Learned SR1 Preconditioning (ICML 2026)

Independent reproduction of ICML 2026 paper w1fkDwiZgN (arXiv 2508.12270): logbook, models, and analytic-benchmark artifacts.

Running

Repro - L-SR1: Learned Symmetric-Rank-One Preconditioning

🎯

Explore and manage experiment logs in an interactive web logbook
txus/lsr1-artifacts

10.9 MB

Repro - BALLAST (ICML 2026, 0xOj6kVMbb)

From-scratch reproduction of BALLAST (arXiv 2509.26005): 3/5 claims reproduced, 1 partial, 1 not attempted.

Running

Repro - BALLAST: Bayesian Active Learning with Look-ahead Amendment for Sea-drifter Trajectories under Spatio-Temporal Vector Fields

🎯

Explore research logs and collaborate with an AI agent
txus/ballast-repro

Updated 1 day ago • 72
txus/ballast-repro-results

Updated 1 day ago • 48

fSGLD ICML-2026 Reproduction (Kw0nhxeTeI)

Reproduction of Flatness-Aware SGLD (arXiv 2510.02174, ICML 2026): theory + CIFAR-100N experiments.

txus/fsgld-cifar100n-repro

Viewer • Updated 2 days ago • 10 • 41
Running

Repro - Flatness-Aware Stochastic Gradient Langevin Dynamics

🎯

Log and sync research notes with an AI coding agent

Repro: Epistemic Uncertainty Overparam NNs

Reproduction of Ruegamer, ICML 2026 (arXiv 2605.25234): all 4 claims. Trackio logbook + bundle.

Running

Repro - On the Epistemic Uncertainty of Overparametrized Neural Networks

🎯

Browse and sync research logbook with your coding agent

Reproduction: Rare Event Analysis of LLMs (ICML 2026 #71154)

Independent reproduction of arXiv 2602.06791 (Dorman et al.). MBAR/TPS rare-event estimation for TinyStories-8M, ARI & Log-Prob observables.

txus/rea-llm-repro-71154

Updated about 17 hours ago
Running

Repro: Rare Event Analysis of Large Language Models

🎯

Collaborate on LLM rare-event logs with a coding agent
Rare Event Analysis of Large Language Models

Paper • 2602.06791 • Published Feb 6

Repro - BALLAST (ICML 2026, 0xOj6kVMbb)

From-scratch reproduction of BALLAST (arXiv 2509.26005): 3/5 claims reproduced, 1 partial, 1 not attempted.

Running

Repro - BALLAST: Bayesian Active Learning with Look-ahead Amendment for Sea-drifter Trajectories under Spatio-Temporal Vector Fields

🎯

Explore research logs and collaborate with an AI agent
txus/ballast-repro

Updated 1 day ago • 72
txus/ballast-repro-results

Updated 1 day ago • 48

Repro: Janus-LoRA (ICML 2026, XJ5r19iN9G)

Reproduction of Janus-LoRA — all 6 claims. Logbook Space, GPU-run dataset, and artifact bucket.

Running

Repro - Janus-LoRA: A Balanced Low-Rank Adaptation for Continual Learning

⚖

Explore experiment logs and sync them with a coding agent
txus/janus-lora-repro

Updated 2 days ago • 48

fSGLD ICML-2026 Reproduction (Kw0nhxeTeI)

Reproduction of Flatness-Aware SGLD (arXiv 2510.02174, ICML 2026): theory + CIFAR-100N experiments.

txus/fsgld-cifar100n-repro

Viewer • Updated 2 days ago • 10 • 41
Running

Repro - Flatness-Aware Stochastic Gradient Langevin Dynamics

🎯

Log and sync research notes with an AI coding agent

Reproduction: L-SR1 Learned SR1 Preconditioning (ICML 2026)

Independent reproduction of ICML 2026 paper w1fkDwiZgN (arXiv 2508.12270): logbook, models, and analytic-benchmark artifacts.

Running

Repro - L-SR1: Learned Symmetric-Rank-One Preconditioning

🎯

Explore and manage experiment logs in an interactive web logbook
txus/lsr1-artifacts

10.9 MB

Repro: Epistemic Uncertainty Overparam NNs

Reproduction of Ruegamer, ICML 2026 (arXiv 2605.25234): all 4 claims. Trackio logbook + bundle.

Running

Repro - On the Epistemic Uncertainty of Overparametrized Neural Networks

🎯

Browse and sync research logbook with your coding agent

Txus Bach PRO

AI & ML interests

Recent Activity

Organizations

txus 's collections 6

Repro: Rare Event Analysis of Large Language Models

Repro - Janus-LoRA: A Balanced Low-Rank Adaptation for Continual Learning

Repro - L-SR1: Learned Symmetric-Rank-One Preconditioning

txus/lsr1-artifacts

Repro - BALLAST: Bayesian Active Learning with Look-ahead Amendment for Sea-drifter Trajectories under Spatio-Temporal Vector Fields

Repro - Flatness-Aware Stochastic Gradient Langevin Dynamics

Repro - On the Epistemic Uncertainty of Overparametrized Neural Networks

Repro: Rare Event Analysis of Large Language Models

Repro - BALLAST: Bayesian Active Learning with Look-ahead Amendment for Sea-drifter Trajectories under Spatio-Temporal Vector Fields

Repro - Janus-LoRA: A Balanced Low-Rank Adaptation for Continual Learning

Repro - Flatness-Aware Stochastic Gradient Langevin Dynamics

Repro - L-SR1: Learned Symmetric-Rank-One Preconditioning

txus/lsr1-artifacts

Repro - On the Epistemic Uncertainty of Overparametrized Neural Networks