metadata

base_model:
  - google/t5-v1_1-xxl
datasets:
  - allenai/tulu-v2-sft-mixture
language:
  - en
library_name: transformers
license: mit
pipeline_tag: text-generation
tags:
  - language-modeling
  - causal-lm
  - bias-analysis
  - cognitive-bias
metrics:
  - accuracy

Model Card for T5-Tulu

Model Details

Model Description
This 🤗 Transformers model was finetuned using LoRA adapters for the arXiv paper:
"Planted in Pretraining, Swayed by Finetuning: A Case Study on the Origins of Cognitive Biases in LLMs"
We study whether cognitive biases in LLMs emerge from pretraining, instruction tuning, or training randomness. This is one of 3 identical versions trained with different random seeds.

Model type: encoder-decoder based transformer
Language(s): English
License: MIT
Finetuned from: google/t5-v1_1-xxl
Paper: https://arxiv.org/abs/2507.07186
Project Page: https://itay1itzhak.github.io/planted-in-pretraining
Repository: https://github.com/itay1itzhak/planted-in-pretraining

Uses

Direct Use

For research on cognitive biases in LLMs. Used to test causal impact of pretraining vs instruction tuning.

Out-of-Scope Use

Do not use in production, sensitive domains, or decision-critical applications.

How to Get Started with the Model

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("itay1itzhak/T5-Tulu")
tokenizer = AutoTokenizer.from_pretrained("itay1itzhak/T5-Tulu")

inputs = tokenizer("Example input?", return_tensors="pt")
outputs = model.generate(**inputs)
print(tokenizer.decode(outputs[0]))

Training Details

Finetuning method: LoRA (high-rank, rank ∈ [64, 512])
Instruction data: Tulu-2
Seeds: 3 per setting to evaluate randomness effects
Batch size: 128 (OLMo) / 64 (T5)
Learning rate: 1e-6 to 1e-3
Steps: ~5.5k (OLMo) / ~16k (T5)
Mixed precision: fp16 (OLMo) / bf16 (T5)

Evaluation

Evaluated on 32 cognitive biases from Itzhak et al. (2024) and Malberg et al. (2024)
Metrics: mean bias score, PCA clustering, MMLU accuracy
Findings: Biases primarily originate in pretraining; randomness introduces moderate variation

Environmental Impact

Hardware: 4× NVIDIA A40
Estimated time: ~120 GPU hours/model

Technical Specifications

Architecture: T5-11B
Instruction dataset: Tulu-2