testing again

.idea/.gitignore

@@ -0,0 +1,13 @@
+# Default ignored files
+/shelf/
+/workspace.xml
+# Editor-based HTTP Client requests
+/httpRequests/
+# Datasource local storage ignored files
+/dataSources/
+/dataSources.local.xml
+/aws.xml
+/misc.xml
+/modules.xml
+/vcs.xml
+/xindus_t5base.iml

app.py

@@ -0,0 +1,130 @@
+import os
+import gradio as gr
+import pandas as pd
+from datasets import Dataset
+from transformers import (
+    T5Tokenizer,
+    T5ForConditionalGeneration,
+    Trainer,
+    TrainingArguments
+)
+from huggingface_hub import login
+from spaces import GPU  # Required for ZeroGPU Spaces
+
+import transformers
+print("🔥 Transformers version:", transformers.__version__)
+
+import torch
+device = torch.device("cpu")
+
+login(token=os.getenv("HF_TOKEN"))  # Using environment variable
+# Disable CUDA and set environment variables for Spaces with Stateless GPU
+os.environ["CUDA_VISIBLE_DEVICES"] = ""  # Ensure no GPUs are used
+os.environ["ACCELERATE_DISABLE"] = "true"  # Disable Accelerate for safety
+
+model_name = "madankn/xindus_t5base"
+
+# Load tokenizer and model once
+tokenizer = T5Tokenizer.from_pretrained("t5-base")
+model = T5ForConditionalGeneration.from_pretrained("t5-base")
+model.to("cpu")  # Ensure model stays on CPU in main process
+
+def preprocess(example):
+    inputs = tokenizer(
+        example["text"],
+        padding="max_length",
+        truncation=True,
+        max_length=512,
+    )
+    labels = tokenizer(
+        example["summary"],
+        padding="max_length",
+        truncation=True,
+        max_length=128,
+    )
+    return {
+        "input_ids": inputs["input_ids"],
+        "attention_mask": inputs["attention_mask"],
+        "labels": labels["input_ids"]
+    }
+
+# Load fine-tuned model for inference
+def load_model():
+    global tokenizer, model
+    if tokenizer is None or model is None:
+        tokenizer = T5Tokenizer.from_pretrained("./fine_tuned_t5")
+        model = T5ForConditionalGeneration.from_pretrained("./fine_tuned_t5")
+        model.to("cpu")
+    return model, tokenizer
+
+def train_model():
+    # 🔁 Reload model/tokenizer inside training function to avoid stale GPU bindings
+    tokenizer = T5Tokenizer.from_pretrained("t5-base")
+    model = T5ForConditionalGeneration.from_pretrained("t5-base").to(device)
+
+    df = pd.read_csv("xindus_dataset.csv")
+    df = df.rename(columns={"text_column_name": "text", "summary_column_name": "summary"})
+    dataset = Dataset.from_pandas(df).train_test_split(test_size=0.1)
+
+    tokenized_datasets = dataset.map(
+        preprocess,
+        batched=True,
+        remove_columns=dataset["train"].column_names
+    )
+
+    training_args = TrainingArguments(
+        output_dir="./results",
+        logging_dir="./logs",
+        logging_steps=50,
+        save_steps=200,
+        num_train_epochs=1,
+        per_device_train_batch_size=2,
+        per_device_eval_batch_size=2,
+        weight_decay=0.01,
+        learning_rate=2e-5,
+        save_total_limit=1,
+        push_to_hub=True,
+        hub_model_id=model_name,
+        hub_strategy="every_save",
+        no_cuda=True  # 🧠 Critical: disable all CUDA use
+    )
+
+    trainer = Trainer(
+        model=model,
+        args=training_args,
+        train_dataset=tokenized_datasets["train"],
+        eval_dataset=tokenized_datasets["test"],
+        tokenizer=tokenizer,
+    )
+
+    trainer.train()
+    model.save_pretrained("./fine_tuned_t5")
+    tokenizer.save_pretrained("./fine_tuned_t5")
+    model.push_to_hub(model_name)
+    tokenizer.push_to_hub(model_name)
+
+    return f"✅ Training complete and pushed to: https://huggingface.co/{model_name}"
+
+# Summarize function using GPU
+@GPU
+def summarize(text):
+    model, tokenizer = load_model()
+    input_ids = tokenizer("summarize: " + text, return_tensors="pt", truncation=True).input_ids
+    output_ids = model.generate(input_ids, max_length=50)
+    return tokenizer.decode(output_ids[0], skip_special_tokens=True)
+
+# Gradio UI
+train_button = gr.Interface(fn=train_model, inputs=[], outputs="text", title="Train T5 on Xindus Data")
+summarize_interface = gr.Interface(fn=summarize, inputs="text", outputs="text", title="Summarize with Fine-Tuned T5")
+
+# Combine interfaces
+def combined_interface():
+    with gr.Blocks() as demo:
+        with gr.Tab("Training"):
+            train_button.render()
+        with gr.Tab("Summarization"):
+            summarize_interface.render()
+    demo.launch()
+
+# Launch app
+combined_interface()

requirements.txt

@@ -0,0 +1,24 @@
+# Core libraries
+transformers>=4.38.0
+datasets>=2.18.0
+pandas>=1.5.3
+
+# For tokenization and model loading
+sentencepiece>=0.1.99
+protobuf<4.0.0  # Avoid some compatibility issues with tokenizers
+
+# Gradio interface
+gradio>=4.14.0
+
+# Hugging Face Hub access
+huggingface_hub>=0.20.0
+
+# For GPU / inference in Spaces (ZeroGPU, if needed)
+spaces==0.22.0  # or latest version if you're using @GPU
+
+# Optional logging and evaluation
+scikit-learn>=1.2.2
+
+# Misc
+numpy>=1.23.0
+accelerate>=0.21.0

xindus_dataset.csv

@@ -0,0 +1,24 @@
+text,summary
+BL2194 Muesli Brown with Asphalt Black Diamond Work with Cotton Thread Heavy Embroidered Bandhgala Designer Indo-Western Blazer, Designer Cotton Blazer Embroidered Brown Black Diamond Work
+7083 Mischka Gray and Mandys Brown Printed with Bright White Sleeves Super Soft Premium Cotton Designer Shirt,Premium Cotton Shirt Designer Gray Mandys Brown White Sleeves
+BL2158 Gunmetal Navy Blue with Jade Black Chintz Textured Cross Buttoned Bandhgala Designer Blazer,Designer Blazer Navy Blue Black Chintz Textured Cross Buttoned
+7942 Wineberry With Tangaroa Navy Blue Plaid and Striped Twill Premium Cotton Designer Shirt,Striped Cotton Designer Shirt Wineberry Tangaroa Navy Blue Plaid
+SOC001 Pack of 5: Sunny Yellow with polka dot Deep Teal Blue Striped Navy Blue And Black Premium Combed Cotton Ankle Length Socks,Cotton Ankle Length Socks Polka Dot Stripped Yellow Blue Black
+BL2165 Desert Storm Cream with Eminence Blue Leaves Printed Premium Cotton Designer Blazer,Cotton designer blazer blue leaves print premium quality
+7825 Bright White with Tuna Blue Plaid Premium Cotton Designer Shirt,Cotton designer shirt bright white with tuna blue plaid
+5834 Bright Turquoise Blue and Dark Cerulean Twill Plaid Premium Cotton Shirt,Cotton plaid shirt in turquoise and cerulean colors
+Jade Black And Bright White Super Soft Premium Cotton Designer Lounge Pant,Cotton lounge pants jade black and white super soft
+8304 Bright White with potters Clay Brown Plaid Flannel Designer Overshirt,Flannel overshirt bright white and clay brown plaid
+Jade Black with White Striped Super Soft Premium Cotton Designer Shirt,Cotton designer shirt jade black with white stripes
+6940 Iroko Brown with Spindle Blue Floral Printed Super Soft Premium Cotton Designer Shirt,Cotton designer shirt Iroko brown blue floral print
+8940 Bright White with Cadet Pink Multicolour Printed Super Soft Premium Cotton Shirt,Giza cotton shirt bright white with jade black patch
+8940 Bright White with Cadet Pink Multicolour Printed Super Soft Premium Cotton Shirt,Cotton shirt bright white with cadet pink print
+5622 Pacific Blue Striped Dobby Textured Premium Giza Cotton Shirt,Giza cotton striped shirt premium quality Pacific Blue
+8940 Bright White with Cadet Pink Multicolour Printed Super Soft Premium Cotton Shirt,Cotton shirt bright white with cadet pink print
+ST460CBG Navy Cross Buttoned Bandhgala/Mandarin Wool-Silk blend Suit,Wool-silk blend Navy Bandhgala suit with cross buttons
+8166 Half Charcoal Grey and Half Catalina Blue Chambray Textured Premium Cotton Hoodie Designer jacket,Cotton hoodie half charcoal grey half blue chambray
+6174 Cloud Burst Blue Butterfly Printed Premium Cotton Kurta Shirt,Cotton kurta shirt blue butterfly print premium quality
+6834 Manatee Grey with Multicolour Floral Print Premium Cotton Shirt,Cotton shirt Manatee Grey multicolour floral print
+TS524 Manatee with Cloud Burst Jacquard Textured Super Soft Pique T-Shirt,Manatee with Cloud Burst Jacquard Textured Super Soft Pique T-Shirt
+
+