alibidaran/Platio_merged_model

PlaiTO 🧠✨

A Reasoning-Focused Language Model for the Humanities

Overview

PlaiTO is a reasoning-oriented language model designed specifically for humanities and social sciences. Built on top of LLaMA 3.1 8B, PlaiTO emphasizes structured thinking, conceptual understanding, and analytical reasoning rather than surface-level text generation.

The model performs especially well in domains where theory, interpretation, decision-making, and human behavior matter most.

Base Model

• Architecture: LLaMA 3.1
• Parameters: 8B
• Training Focus: Reasoning, conceptual analysis, and humanities-oriented problem solving

Target Domains

PlaiTO is optimized for:

• Psychology
• Management & Organizational Studies
• Sociology
• Related humanities and social science disciplines

Typical use cases include:

• Theoretical analysis
• Case study reasoning
• Concept explanation and comparison
• Decision-making support
• Academic discussion and synthesis

Benchmark Performance

PlaiTO was evaluated on the MMLU benchmark using 100 samples per subject area. Results show strong and consistent performance across key humanities domains:

Domain	Accuracy
Professional Psychology	76%
Management	74%
Sociology	75%

These results indicate reliable reasoning capabilities in complex, abstract, and theory-heavy tasks.

Strengths

• Strong reasoning and analytical depth
• Better handling of abstract concepts and human-centered problems
• Suitable for academic, educational, and research-oriented applications
• Balanced performance across multiple humanities disciplines

Limitations

• Not optimized for mathematics-heavy or symbolic reasoning tasks
• May underperform in domains requiring exact numerical computation
• As with all LLMs, outputs should be reviewed for accuracy in high-stakes settings

Intended Use

PlaiTO is intended for:

• Research and academic exploration
• Educational tools and tutoring systems
• Decision-support in management and organizational contexts
• Exploratory analysis in psychology and sociology

Direct Usage

from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
from peft import PeftModel
import torch

# Define the model IDs
base_model_name_or_path = "alibidaran/Platio_merged_model" # The base Llama-3-8B-Instruct model

# 1. Configure 4-bit quantization
bnb_config = BitsAndBytesConfig(
   load_in_4bit=True,
   bnb_4bit_use_double_quant=True,
   bnb_4bit_quant_type="nf4",
   bnb_4bit_compute_dtype=torch.float16
)

# 2. Load the Base Model with the config
# Use device_map="auto" for efficient loading with quantization
# Use torch_dtype=torch.bfloat16 for Llama models with bnb
model = AutoModelForCausalLM.from_pretrained(
   base_model_name_or_path,# The PEFT adapter ID
   quantization_config=bnb_config,
   torch_dtype=torch.bfloat16,
   device_map="cuda",
)

tokenizer=AutoTokenizer.from_pretrained(base_model_name_or_path)
system_prompt="""
   You are a reasonable expert who thinks and answer the users question.
   Before respond first think and create a chain of thoughts in your mind.
   Then respond to the client.
   Your chain of thought and reflection must be in <thinking>..</thinking> format and your respond
   should be in the <output>..</output> format.
   """
  
   messages = [
               {'role':'system','content':system_prompt},
               {"role": "user", "content":message},

               ] 
     
   inputs = tokenizer.apply_chat_template(
           messages,
       tokenize = True,
       add_generation_prompt = True, # Must add for generation
       return_tensors = "pt",).to("cuda")
   inputs_shape=inputs['input_ids'].shape[1]
   with torch.no_grad():
       output=model.generate(**inputs, max_new_tokens =2048,
                  use_cache = True, temperature = 0.5, min_p = 0.9)

Ethical Considerations

While PlaiTO is designed to reason about human behavior and society, it should not be used as a replacement for professional judgment in clinical, legal, or organizational decision-making. Always apply human oversight.

License

Please refer to the license of the base LLaMA 3.1 model and ensure compliance with its terms.