ReAct reproduction

Reproduction of ICLR 2023 paper "ReAct: Synergizing Reasoning and Acting in Language Models". Developed as part of an application for EEML 2025, Sarajevo, Bosnia and Herzegovina.

⚡Quickstart

Prerequisites

• Python 3.9+
• Conda (for environment management)

🛠️Setup

Clone the repository and install dependencies:

git clone https://github.com/AStroCvijo/react_reproduction.git
cd react_reproduction
conda create --name react python=3.9
conda activate react
conda install -c conda-forge libstdcxx-ng
pip install -r requirements.txt

---

🖥️ Scripts for Running Experiments

🔍 FEVER Dataset

Name	Description	Command
Standard	Standard inference (no reasoning/acting)	./scripts/fever/standard.sh
CoT	Chain-of-Thought (CoT)	./scripts/fever/cot.sh
CoT-SC	CoT with self-consistency (21 samples)	./scripts/fever/cot_sc.sh
Act	Action-only (no reasoning)	./scripts/fever/act.sh
ReAct	ReAct (reasoning + acting)	./scripts/fever/react.sh
CoT-SC -> ReAct	CoT with self-consistency and ReAct hybrid	./scripts/fever/cot_sc_react.sh
ReAct -> CoT-SC	ReAct and CoT with self-consistency hybrid	./scripts/fever/react_cot_sc.sh

🍲 HotpotQA Dataset

Name	Description	Command
Standard	Standard inference (no reasoning/acting)	./scripts/hotpotqa/standard.sh
CoT	Chain-of-Thought (CoT)	./scripts/hotpotqa/cot.sh
CoT-SC	CoT with self-consistency (21 samples)	./scripts/hotpotqa/cot_sc.sh
Act	Action-only (no reasoning)	./scripts/hotpotqa/act.sh
ReAct	ReAct (reasoning + acting)	./scripts/hotpotqa/react.sh
CoT-SC -> ReAct	CoT with self-consistency and ReAct hybrid	./scripts/hotpotqa/cot_sc_react.sh
ReAct -> CoT-SC	ReAct and CoT with self-consistency hybrid	./scripts/hotpotqa/react_cot_sc.sh

🏠 ALFWorld Dataset

Name	Description	Command
Act	Action-only (no reasoning)	./scripts/alfworld/act.sh
ReAct	ReAct (reasoning + acting)	./scripts/alfworld/react.sh

🛍️ WebShop Dataset

Name	Description	Command
Act	Action-only (no reasoning)	./scripts/webshop/act.sh
ReAct	ReAct (reasoning + acting)	./scripts/webshop/react.sh

---

📖Arguments Guide

Argument	Description	Default	Options
-ds, --data_set	Dataset selection	FEVER	FEVER, HotpotQA, ALFWorld, WebShop
-ps, --prompt_style	Prompt style to use	ReAct	ReAct, Act, CoT, Standard, CoT-SC-ReAct, ReAct-CoT-SC
-ns, --num_samples	Number of samples to generate	1	Any positive integer
-t, --tempreture	Temperature setting for response variability	0.0	Any float value (0.0 to 1.0)

📄Experiments

For detailed experiments and evaluations, please refer to the following document:

Experiments and Evaluations