Quickstart

Train a character-level GPT from scratch and generate text in under a minute.

---

1. Train on Names

from strands_microgpt import MicroGPT

# Load dataset → build tokenizer → create model (one line)
model, tokenizer, docs = MicroGPT.from_dataset()

# Train (1000 steps on 32K names)
model.train_on_docs(docs, tokenizer, num_steps=1000)

# Generate
for name in model.generate(tokenizer, num_samples=10):
    print(name)

Expected output (after ~60s of training):

 1. kayla
 2. maren
 3. joline
 4. arian
 5. delia

---

2. Use as a Strands Agent

from strands import Agent
from strands_microgpt import MicroGPTModel

model = MicroGPTModel(num_steps=500, temperature=0.5)
agent = Agent(model=model)
agent("Generate some names")

The model trains on first use, then generates text for every agent call.

---

3. Use as a Tool

from strands import Agent
from strands_microgpt import microgpt_train, microgpt_generate

# Works with any model provider (Bedrock, OpenAI, etc.)
agent = Agent(tools=[microgpt_train, microgpt_generate])

agent("Train a MicroGPT on names for 500 steps")
agent("Generate 10 names with temperature 0.7")

---

4. Save & Load Checkpoints

# Save after training
model.save_checkpoint("my_model.json", tokenizer)

# Load and generate later
model, tokenizer, meta = MicroGPT.load_checkpoint("my_model.json")
samples = model.generate(tokenizer, num_samples=20)

---

Configuration

Parameter	Default	Description
n_layer	1	Transformer depth
n_embd	16	Embedding dimension
block_size	16	Context window
n_head	4	Attention heads
num_steps	1000	Training steps
learning_rate	0.01	Initial learning rate
temperature	0.5	Sampling temperature

→ Next: Autograd Engine | Custom Datasets