Robot

Democratizing AI via Software Solutions

Train and run high performing, large-scale models on commodity platforms including desktops and laptops.

Try it now

Train with Sparsity


from thirdai import bolt 

layers = [
  bolt.FullyConnected(
    dim = 2000,
    load_factor = 0.05, # Sparsity: Use only 5% FLOPs 
    activation_function = bolt.ActivationFunctions.ReLU),

  bolt.FullyConnected(
    dim = 2,
    load_factor = 1.0,
    activation_function = bolt.ActivationFunctions.Softmax),
]

network = bolt.Network(
  layers = layers,
  dim = 100000
)

Product Search


from thirdai import DocRetrieval
from thirdai import DocSearchModel

embedding_model = DocSearchModel()
our_index = DocRetrieval(
	dense_input_dimension=128, 
  	num_tables=16, 
  	hashes_per_table=6, 
  	centroids=reduced_centroids)

for doc_id, doc_text in dataset:
	embedding = embedding_model.encodeDocs([doc_text])
	our_index.add_doc(doc_id=doc_id, doc_text=doc_text, doc_embeddings=embedding)

while True:
	query_text = input()
	embedding = embedding_model.encodeQuery(query_text)
	result = our_index.query(embedding, top_k=1)
	print(result[0])

Sentiment Classification


network = bolt.Network(
  layers = layers,
  dim = 100000)

network.train(
	train_data=train_data,
	loss_fn=bolt.CategoricalCrossEntropyLoss(), 
	learning_rate=0.0001, 
	epochs=20, 
	verbose=True)

predictions, metric_results = network.predict(
	test_data=test_data, 
	metrics=["categorical_accuracy"], 
	verbose=True)

print(predictions)
print(metric_results)

Introducing The Big OL' Layer Training (Bolt) Engine

ThirdAI is pleased to provide 60 day trial release our BOLT Engine via PyPI. Our software is ready to be installed on any machine  for fast training of deep learning models on any CPU. With BOLT, old refurbished desktops can be more powerful than top-of-the-line GPUs, capable of training models with hundreds of millions to billions of parameters.

Learn More

Introducing ThirdAI’s BOLT Engine

AI models have grown tremendously in the past few years. We have reached a stage where training one AI model can emit as much carbon as five…

Read Article

Got CPU? Unlocking AI Speed and Flexibility...

The current AI industry assumes that CPUs are inferior to GPUs and other specialized processors (such as TPUs) for heavyweight AI computations.

Read Article
Shrivastava

Shrivastava’s AI startup receives $6M in...

Rice Computer Science associate professor Anshumali Shrivastava’s start-up company, Houston-based ThirdAI has received…

Read Article

Why Cervin Ventures Invested in ThirdAI

ThirdAI, a cutting-edge Artificial Intelligence startup building tools to speed up deep learning without the need for specialized hardware…

Read Article