app.py

from transformers import pipeline, AutoModelForCausalLM, AutoTokenizer
import random
import gradio as gr

# Model options
model_options = {
    "distilgpt2": "distilgpt2",
    "GPT-Neo 125M": "EleutherAI/gpt-neo-125M",
}

# Load default model
default_model_name = model_options["GPT-Neo 125M"]
tokenizer = AutoTokenizer.from_pretrained(default_model_name)
model = AutoModelForCausalLM.from_pretrained(default_model_name)
generator = pipeline("text-generation", model=model, tokenizer=tokenizer, device=-1)  # Use CPU

# Predefined options for randomization
names = ["John Doe", "Jane Smith", "Ali Khan"]
locations = ["Pump House 1", "Main Valve Station", "Chemical Storage Area"]
work_types = ["Routine pump maintenance", "Valve inspection", "Chemical handling"]
durations = [30, 45, 60]
good_practices = ["Good Practice"]
deviations = ["Deviation"]

plant_observations = [
    ("Energy sources controlled", "Good Practice", "Lockout/tagout procedures were followed."),
    ("Leaks/spills contained", "Deviation", "Oil spill near a pump flagged for cleanup."),
    ("Housekeeping standard high", "Deviation", "Scattered tools were organized after reminder."),
]

# Function to set seed
def set_seed(seed_value):
    random.seed(seed_value)

# AI-based SOC report generation
def generate_soc(model_choice, seed=None):
    # Set seed if provided
    if seed:
        set_seed(seed)

    # Update the generator if model_choice changes
    global generator
    model_name = model_options[model_choice]
    if generator.tokenizer.name_or_path != model_name:
        tokenizer = AutoTokenizer.from_pretrained(model_name)
        model = AutoModelForCausalLM.from_pretrained(model_name)
        generator = pipeline("text-generation", model=model, tokenizer=tokenizer, device=-1)

    # Randomized fields
    observer_name = random.choice(names)
    location = random.choice(locations)
    work_type = random.choice(work_types)
    duration = random.choice(durations)

    # Generate random plant observations
    observations = "\n".join(
        f"{i+1}. {obs[0]}\n{obs[1]}\n{obs[2]}"
        for i, obs in enumerate(random.sample(plant_observations, len(plant_observations)))
    )

    # AI Prompt
    prompt = f"""
    Write a detailed Safety Observation and Conversation (SOC) report for a water injection plant. 

    Key Safety Conclusions/Comments/Agreements Made:
    Briefly summarize safety observations, key concerns, and corrective actions.

    Observer's Name: {observer_name}
    KOC ID No.: [Insert KOC ID here]
    Type of Work Observed: {work_type}
    Location: {location}
    Duration (in mins): {duration}

    --- Plant Observations:
    {observations}

    --- People Observations:
    Include details on PPE compliance, hazard understanding, and good practices or deviations.

    --- Process Observations:
    Summarize job safety analysis, procedures followed, and improvements needed.

    --- Performance Observations:
    Evaluate the overall safety performance, including work pace and supervision.
    """
    result = generator(prompt, max_length=512, num_return_sequences=1)[0]["generated_text"]
    return result

# Gradio Interface
def app_interface(model_choice, seed):
    return generate_soc(model_choice, seed)

# Gradio Layout
with gr.Blocks() as app:
    gr.Markdown("# AI-Generated Safety Observation and Conversation (SOC) Reports")
    gr.Markdown(
        """
        Generate detailed SOC reports for a water injection plant using AI assistance. 
        Customize your report with multiple models, randomization, and reproducibility through seeds.
        """
    )

    with gr.Row():
        model_choice = gr.Dropdown(
            label="Select Model",
            choices=list(model_options.keys()),
            value="GPT-Neo 125M",
        )
        seed = gr.Number(label="Seed (Optional)", value=None, precision=0)

    output_box = gr.Textbox(
        label="Generated SOC Report",
        placeholder="Your SOC report will appear here...",
        lines=30,
    )

    with gr.Row():
        generate_btn = gr.Button("Generate SOC Report")
        copy_btn = gr.Button("Copy to Clipboard")

    generate_btn.click(app_interface, inputs=[model_choice, seed], outputs=output_box)
    copy_btn.click(lambda text: text, inputs=output_box, outputs=None)

# Launch the app
app.launch()