---
library_name: transformers
tags: [PRefLexOR]
---

# PRefLexOR: Preference-based Recursive Language Modeling for Exploratory Optimization of Reasoning and Agentic Thinking

We introduce PRefLexOR (Preference-based Recursive Language Modeling for Exploratory Optimization of Reasoning), a framework that combines preference optimization with concepts from Reinforcement Learning (RL) to enable models to self-teach through iterative reasoning improvements. Central to PRefLexOR are thinking tokens, which explicitly mark reflective reasoning phases within model outputs, allowing the model to recursively engage in multi-step reasoning, revisiting, and refining intermediate steps before producing a final output. The foundation of PRefLexOR lies in Odds Ratio Preference Optimization (ORPO), where the model learns to align its reasoning with human-preferred decision paths by optimizing the log odds between preferred and non-preferred responses. The integration of Direct Preference Optimization (DPO) further enhances model performance by using rejection sampling to fine-tune reasoning quality, ensuring nuanced preference alignment. This hybrid approach between ORPO and DPO mirrors key aspects of RL, where the model is continuously guided by feedback to improve decision-making and reasoning. Active learning mechanisms allow PRefLexOR to dynamically generate new tasks, reasoning steps, and rejected answers on-the-fly during training. This adaptive process enables the model to self-teach as it continually improves through real-time feedback and recursive processing. 

Our method diverges from traditional approaches by not relying on pre-generated datasets; instead, it dynamically generates new tasks, reasoning steps, and feedback on the fly, allowing the model to continuously adapt and improve in real time. Recursive optimization within the thinking token framework introduces iterative feedback loops, where the model refines its reasoning, much like policy refinement in RL, achieving deeper coherence, consistency, and adaptability. By recursively optimizing reasoning through feedback-driven learning, PRefLexOR achieves significant flexibility in its ability to handle complex tasks, learning and evolving its cognitive abilities autonomously. This framework advances the field of cognitive alignment by demonstrating that models can iteratively teach themselves to reason with greater depth and reflectivity, akin to an RL-based self-improving system capable of solving open-domain problems with superior reasoning depth and logic. Our implementation is straightforward and can be Incorporated into any existing pretrained LLM. The approach is demonstrated in use cases of materials design applications, where a small language model is trained to develop sophisticated reasoning capabilities. Thereby, PRefLexOR builds a dynamic knowledge graph by generating questions from random text and using Retrieval-Augmented Generation (RAG) to retrieve contextually relevant data from the entire corpus, facilitating recursive reasoning through complex interactions between similar nodes in the embedding space.

Source code: [https://github.com/lamm-mit/PRefLexOR](https://github.com/lamm-mit/PRefLexOR)

![Fig_100](https://github.com/user-attachments/assets/800de09d-64c4-4ead-903f-80525f8bf415)

Figure 1: Illustration of the workflow and design principles behind generative materials informatics. Panel a: The process of transforming information into knowledge and actionable outcomes. Each individual piece of information (left) is synthesized into a network of interconnected knowledge, leading to informed decisions and innovative designs (right). Panel b: Conventional approaches in materials science rely on data-driven models, partial differential equations (PDEs), and experimental results, focusing on single-step predictions. Panel c: In contrast, generative materials informatics models built on the PRefLexOR framework proposed in this paper use 'thinking' and 'reflection' explicitly by incorporating iterative reasoning and contextual understanding, allowing for more complex, multi-step predictions. This approach expands from single inference steps, includes multiple modalities of data and responses, integrates real-world feedback and physics, and leverages self-assessment and self-learning. Using using reinforcement learning (RL) principles, the discovery of principles or the solution of specific tasks is further inspired by biological paradigms, using bio-inspired neural network designs. These advanced methods support continuous improvement in material predictions, enabling more adaptable and intelligent designs

![image](https://github.com/user-attachments/assets/1119b9f7-5f45-4712-81a5-11699a02c571)

Figure 2: PRefLexOR Recursive Reasoning Algorithm: An iterative approach leveraging a fine-tuned Reasoning Model and a general-purpose Critic Model to generate, refine, and optionally integrate responses. The process involves generating initial responses, extracting reflections, improving thinking processes, and creating new responses based on refined thinking, with an optional final integration step. The algorithm relies on extracting thinking processes (indicated via ```<|thinking|>...<|/thinking|>```) and reflection processes  (indicated via ```<|reflect|>...<|/reflect|>```). The use of special tokens allows us to easily construct such agentic modeling as it facilitates pausing inference, improving the strategy, and re-generating improved answers. The sampled responses can either be used in their final state or integrated into an amalgamated response that shows very rich facets in the scientific process.  


```python
from transformers import AutoModelForCausalLM, AutoTokenizer

model_name='lamm-mit/PRefLexOR_ORPO_DPO_EXO_10242024'
model = AutoModelForCausalLM.from_pretrained(model_name,     
    torch_dtype =torch.bfloat16,
    attn_implementation="flash_attention_2",device_map="auto",trust_remote_code=True,
    )
tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True,
                                          use_fast=False,
                                         )
```

## Inference example

###  Simple inference:

```python
from PRefLexOR import *

txt = 'What is the relationship between materials and music? Brief answer.' + f' Use {think_start}.'

output_text, messages = generate_local_model(
    model=model, 
    tokenizer=tokenizer, 
    prompt=txt, 
    system_prompt='',  
    num_return_sequences=1, 
    repetition_penalty=1.0, 
    temperature=0.1, 
    max_new_tokens=2024, 
    messages=[], 
    do_sample=True
)

print(output_text)
```
Extract thinking and output:

```python
thinking    = extract_text(output_text, thinking_start=think_start, thinking_end=think_end)[0].strip()
answer_only = extract_text(output_text, thinking_start=think_end, thinking_end="NONE").strip()
print ("THINKING:\n\n", thinking)
print ("ANSWER:\n\n", answer_only)
```

### Recursive inference usingh multi-agentic modeling 

```python
from transformers import AutoModelForCausalLM, AutoTokenizer

# Load reasoning model
model_name='lamm-mit/PRefLexOR_ORPO_DPO_EXO_10242024'
model = AutoModelForCausalLM.from_pretrained(model_name,     
    torch_dtype =torch.bfloat16,
    attn_implementation="flash_attention_2",device_map="auto",trust_remote_code=True,
    )
tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True,
                                          use_fast=False,
                                         )
# Load critic model
model_name_critic = "meta-llama/Llama-3.2-3B-Instruct"

critic_model = AutoModelForCausalLM.from_pretrained(
    model_name_critic, 
    torch_dtype=torch.bfloat16, 
    attn_implementation="flash_attention_2", 
    device_map="auto", 
    trust_remote_code=True
)
```

Example inference 
```python
output_text, output_list, output_text_integrated = recursive_response_from_thinking(
    model=model, 
    tokenizer=tokenizer, 
    model_critic=critic_model, 
    tokenizer_critic=tokenizer,  #same tokenizer in our case
    question="Develop an idea of how graphene can be combined with silk fibers to create a filtration membrane.", 
    N=3, 
    temperature=0.1, 
    temperature_improvement=0.1, 
    system_prompt="You are a helpful assistant.", 
    system_prompt_critic="You carefully improve responses, with attention to detail, and following all directions.",
    verbatim=False,
```
Printing the output:
```python
for i, item in enumerate(output_list):
    print (f"i={i}", 64*"-")
    print (item)

print (64*"#")
print ("INTEGRATED RESPONSE:")
print (output_text_integrated)
print (64*"#")

```

## Citation 

```bibtex
@article{buehler2024PRefLexOR,
      title={PRefLexOR: Preference-based Recursive Language Modeling for Exploratory Optimization of Reasoning and Agentic Thinking}, 
      author={Markus J. Buehler},
      year={2024},
      eprint={2410.12375},
      archivePrefix={arXiv},
      primaryClass={cs.AI},
      url={https://arxiv.org/abs/2410.12375}, 
}
```