veda-programming / train.py
vedaco's picture
Create train.py
dbb535a verified
raw
history blame
9.66 kB
import tensorflow as tf
from tensorflow import keras
import numpy as np
import os
from typing import List, Tuple, Optional
from model import VedaProgrammingLLM, create_veda_model
from tokenizer import VedaTokenizer
class VedaTrainer:
"""Trainer class for Veda Programming LLM"""
def __init__(
self,
data_path: str = "programming.txt",
vocab_size: int = 10000,
max_length: int = 256,
batch_size: int = 32,
model_size: str = "small"
):
self.data_path = data_path
self.vocab_size = vocab_size
self.max_length = max_length
self.batch_size = batch_size
self.model_size = model_size
self.tokenizer = VedaTokenizer(vocab_size=vocab_size)
self.model: Optional[VedaProgrammingLLM] = None
def load_data(self) -> List[str]:
"""Load programming data from file"""
if not os.path.exists(self.data_path):
print(f"Creating sample {self.data_path}...")
self._create_sample_data()
with open(self.data_path, 'r', encoding='utf-8') as f:
content = f.read()
# Split into code samples (by double newlines or function definitions)
samples = []
current_sample = []
for line in content.split('\n'):
if line.strip() == '' and current_sample:
samples.append('\n'.join(current_sample))
current_sample = []
else:
current_sample.append(line)
if current_sample:
samples.append('\n'.join(current_sample))
# Filter empty samples
samples = [s.strip() for s in samples if s.strip()]
print(f"Loaded {len(samples)} code samples")
return samples
def _create_sample_data(self):
"""Create sample programming data"""
sample_code = '''
def hello_world():
print("Hello, World!")
return True
def fibonacci(n):
if n <= 1:
return n
return fibonacci(n-1) + fibonacci(n-2)
def factorial(n):
if n == 0:
return 1
return n * factorial(n-1)
class Calculator:
def __init__(self):
self.result = 0
def add(self, a, b):
self.result = a + b
return self.result
def subtract(self, a, b):
self.result = a - b
return self.result
def multiply(self, a, b):
self.result = a * b
return self.result
def divide(self, a, b):
if b != 0:
self.result = a / b
return self.result
def bubble_sort(arr):
n = len(arr)
for i in range(n):
for j in range(0, n-i-1):
if arr[j] > arr[j+1]:
arr[j], arr[j+1] = arr[j+1], arr[j]
return arr
def binary_search(arr, target):
left, right = 0, len(arr) - 1
while left <= right:
mid = (left + right) // 2
if arr[mid] == target:
return mid
elif arr[mid] < target:
left = mid + 1
else:
right = mid - 1
return -1
def quicksort(arr):
if len(arr) <= 1:
return arr
pivot = arr[len(arr) // 2]
left = [x for x in arr if x < pivot]
middle = [x for x in arr if x == pivot]
right = [x for x in arr if x > pivot]
return quicksort(left) + middle + quicksort(right)
class LinkedList:
def __init__(self):
self.head = None
def append(self, data):
new_node = Node(data)
if not self.head:
self.head = new_node
return
current = self.head
while current.next:
current = current.next
current.next = new_node
def merge_sort(arr):
if len(arr) <= 1:
return arr
mid = len(arr) // 2
left = merge_sort(arr[:mid])
right = merge_sort(arr[mid:])
return merge(left, right)
def is_palindrome(s):
s = s.lower().replace(" ", "")
return s == s[::-1]
def count_words(text):
words = text.split()
return len(words)
async def fetch_data(url):
async with aiohttp.ClientSession() as session:
async with session.get(url) as response:
return await response.json()
def read_file(filename):
with open(filename, 'r') as f:
return f.read()
def write_file(filename, content):
with open(filename, 'w') as f:
f.write(content)
'''
with open(self.data_path, 'w', encoding='utf-8') as f:
f.write(sample_code)
print(f"Created sample {self.data_path}")
def prepare_dataset(self, samples: List[str]) -> tf.data.Dataset:
"""Prepare TensorFlow dataset for training"""
# Fit tokenizer
self.tokenizer.fit(samples)
# Encode all samples
all_tokens = []
for sample in samples:
tokens = self.tokenizer.encode(sample)
all_tokens.extend(tokens)
# Create sequences
sequences = []
for i in range(0, len(all_tokens) - self.max_length, self.max_length // 2):
seq = all_tokens[i:i + self.max_length + 1]
if len(seq) == self.max_length + 1:
sequences.append(seq)
if not sequences:
# Create padded sequences if not enough data
for sample in samples:
tokens = self.tokenizer.encode(sample, max_length=self.max_length + 1)
sequences.append(tokens)
print(f"Created {len(sequences)} training sequences")
# Convert to numpy arrays
sequences = np.array(sequences)
# Split into input and target
X = sequences[:, :-1]
y = sequences[:, 1:]
# Create dataset
dataset = tf.data.Dataset.from_tensor_slices((X, y))
dataset = dataset.shuffle(buffer_size=len(sequences))
dataset = dataset.batch(self.batch_size)
dataset = dataset.prefetch(tf.data.AUTOTUNE)
return dataset
def build_model(self):
"""Build the Veda Programming model"""
self.model = create_veda_model(
vocab_size=self.tokenizer.vocabulary_size,
max_length=self.max_length,
model_size=self.model_size
)
# Compile model
optimizer = keras.optimizers.Adam(learning_rate=1e-4)
loss_fn = keras.losses.SparseCategoricalCrossentropy(from_logits=True)
self.model.compile(
optimizer=optimizer,
loss=loss_fn,
metrics=['accuracy']
)
# Build model with dummy input
dummy_input = tf.zeros((1, self.max_length), dtype=tf.int32)
self.model(dummy_input)
self.model.summary()
return self.model
def train(
self,
epochs: int = 10,
save_path: str = "veda_model"
):
"""Train the model"""
# Load and prepare data
samples = self.load_data()
dataset = self.prepare_dataset(samples)
# Build model
self.build_model()
# Callbacks
callbacks = [
keras.callbacks.ModelCheckpoint(
filepath=os.path.join(save_path, "model_checkpoint.keras"),
save_best_only=True,
monitor='loss'
),
keras.callbacks.EarlyStopping(
monitor='loss',
patience=5,
restore_best_weights=True
),
keras.callbacks.ReduceLROnPlateau(
monitor='loss',
factor=0.5,
patience=2
)
]
# Create save directory
os.makedirs(save_path, exist_ok=True)
# Train
history = self.model.fit(
dataset,
epochs=epochs,
callbacks=callbacks
)
# Save final model and tokenizer
self.model.save_weights(os.path.join(save_path, "model_weights.h5"))
self.tokenizer.save(os.path.join(save_path, "tokenizer.json"))
# Save model config
config = self.model.get_config()
config['tokenizer_vocab_size'] = self.tokenizer.vocabulary_size
import json
with open(os.path.join(save_path, "config.json"), 'w') as f:
json.dump(config, f)
print(f"Model saved to {save_path}")
return history
def generate(
self,
prompt: str,
max_new_tokens: int = 100,
temperature: float = 0.7
) -> str:
"""Generate code from prompt"""
if self.model is None:
raise ValueError("Model not loaded. Train or load a model first.")
# Encode prompt
prompt_tokens = self.tokenizer.encode(prompt)
# Generate
generated_tokens = self.model.generate(
prompt_tokens,
max_new_tokens=max_new_tokens,
temperature=temperature
)
# Decode
generated_text = self.tokenizer.decode(generated_tokens)
return generated_text
def main():
"""Main training function"""
trainer = VedaTrainer(
data_path="programming.txt",
vocab_size=10000,
max_length=256,
batch_size=16,
model_size="small"
)
# Train model
history = trainer.train(epochs=20, save_path="veda_model")
# Test generation
test_prompt = "def calculate"
generated = trainer.generate(test_prompt, max_new_tokens=50)
print(f"\nGenerated code:\n{generated}")
if __name__ == "__main__":
main()