Training, evaluation datasets and model outputs for the From Leaky Thoughts to Private Reasoning: Controlling What LRMs Say to Themselves
Haritz Puerto
haritzpuerto
AI & ML interests
Reasoning in LLMs, AI safety, agents
Recent Activity
updated a dataset 8 days ago
haritzpuerto/controlling-reasoning-models-privacy-outputs updated a model 27 days ago
haritzpuerto/microsoft-Phi-4-14B-IF-RT updated a model 27 days ago
haritzpuerto/microsoft-Phi-4-14B-IF-FAOrganizations
Models from the ACL 2025 paper "Fine-Tuning on Diverse Reasoning Chains Drives Within-Inference CoT Refinement in LLMs"
"
-
Fine-Tuning with Divergent Chains of Thought Boosts Reasoning Through Self-Correction in Language Models
Paper • 2407.03181 • Published • 1 -
haritzpuerto/LLaMA2-7B-dcot
Text Generation • Updated • 18 • 2 -
haritzpuerto/LLaMA2-13B-dcot
Text Generation • Updated • 11 -
haritzpuerto/LLaMA2-70B-dcot
Text Generation • Updated • 4
Training, evaluation datasets and model outputs for the From Leaky Thoughts to Private Reasoning: Controlling What LRMs Say to Themselves
Training dataset and LoRA checkpoints for the EMNLP 2026 paper From Leaky Thoughts to Private Reasoning: Controlling What LRMs Say to Themselves
Models from the ACL 2025 paper "Fine-Tuning on Diverse Reasoning Chains Drives Within-Inference CoT Refinement in LLMs"
"
-
Fine-Tuning with Divergent Chains of Thought Boosts Reasoning Through Self-Correction in Language Models
Paper • 2407.03181 • Published • 1 -
haritzpuerto/LLaMA2-7B-dcot
Text Generation • Updated • 18 • 2 -
haritzpuerto/LLaMA2-13B-dcot
Text Generation • Updated • 11 -
haritzpuerto/LLaMA2-70B-dcot
Text Generation • Updated • 4
Samples used for the NAACL 2025 Findings paper: "Scaling Up Membership Inference: When and How Attacks Succeed on Large Language Models."