Upload 4 files

app.py

@@ -0,0 +1,17 @@
+import streamlit as st
+from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
+
+st.title("AI Accountant - Prompt-Based ERP Entry")
+
+model_path = "google/flan-t5-large"  # load directly from Hugging Face hub
+
+tokenizer = AutoTokenizer.from_pretrained(model_path)
+model = AutoModelForSeq2SeqLM.from_pretrained(model_path)
+
+user_input = st.text_area("Enter accounting transaction:")
+
+if st.button("Generate Entry"):
+    inputs = tokenizer(user_input, return_tensors="pt")
+    outputs = model.generate(**inputs, max_new_tokens=128)
+    result = tokenizer.decode(outputs[0], skip_special_tokens=True)
+    st.json(eval(result))

requirements.txt

@@ -0,0 +1,4 @@
+transformers
+datasets
+torch
+streamlit

train_data.jsonl

@@ -0,0 +1,5 @@
+{"input": "Today received $245 from Tylor Smith who owed $250 for Samsung X1 sold 3 weeks ago.", "output": "{\"debit\": \"Cash: $245\", \"credit\": \"Accounts Receivable: $245\"}"}
+{"input": "Received inventory worth $400 from ABC LLC on credit.", "output": "{\"debit\": \"Inventory: $400\", \"credit\": \"Accounts Payable: $400\"}"}
+{"input": "Paid $1200 rent for April via bank transfer.", "output": "{\"debit\": \"Rent Expense: $1200\", \"credit\": \"Bank: $1200\"}"}
+{"input": "Sold office chair to John Doe for $300 on credit.", "output": "{\"debit\": \"Accounts Receivable: $300\", \"credit\": \"Sales Revenue: $300\"}"}
+{"input": "Received $3000 from client Smith & Co for past due invoice.", "output": "{\"debit\": \"Cash: $3000\", \"credit\": \"Accounts Receivable: $3000\"}"}

train_flan_t5.py

@@ -0,0 +1,43 @@
+from datasets import load_dataset
+from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, Seq2SeqTrainer, Seq2SeqTrainingArguments
+
+model_checkpoint = "google/flan-t5-large"
+output_dir = "./finetuned-flan-t5"
+
+dataset = load_dataset("json", data_files={"train": "train_data.jsonl"})
+
+tokenizer = AutoTokenizer.from_pretrained(model_checkpoint)
+
+def preprocess_function(examples):
+    inputs = examples["input"]
+    targets = examples["output"]
+    model_inputs = tokenizer(inputs, max_length=512, truncation=True)
+    labels = tokenizer(targets, max_length=128, truncation=True)
+    model_inputs["labels"] = labels["input_ids"]
+    return model_inputs
+
+tokenized_datasets = dataset.map(preprocess_function, batched=True)
+
+model = AutoModelForSeq2SeqLM.from_pretrained(model_checkpoint)
+
+training_args = Seq2SeqTrainingArguments(
+    output_dir=output_dir,
+    evaluation_strategy="no",
+    learning_rate=5e-5,
+    per_device_train_batch_size=2,
+    num_train_epochs=3,
+    weight_decay=0.01,
+    save_total_limit=2,
+    push_to_hub=False
+)
+
+trainer = Seq2SeqTrainer(
+    model=model,
+    args=training_args,
+    train_dataset=tokenized_datasets["train"]
+)
+
+trainer.train()
+
+model.save_pretrained(output_dir)
+tokenizer.save_pretrained(output_dir)