Installation

Get Weaver installed and run your first training in minutes.

System Requirements

• Python: 3.10+
• OS: Linux, macOS, or Windows

Note: The Weaver SDK runs on your local machine (CPU-only is fine). The actual training happens on remote GPU infrastructure that Weaver manages for you.

Weaver SDK Setup

Install the Weaver SDK from PyPI:

pip install nex-weaver

API Key Setup

Before using Weaver, you need to register an account and generate an API key from the Weaver.

Once you have your API key, set it as an environment variable:

export WEAVER_API_KEY=<your-api-key>

You can add this to your .bashrc or .zshrc file to make it permanent.

Verify Installation

Check that Weaver is installed correctly:

python -c "import weaver; print('Weaver installed successfully')"

You should see "Weaver installed successfully" printed.

Quick Start: Your First Training

Let's train a simple Pig Latin translator as a demonstration.

Step 1: Create a Training Script

Create train.py:

import os
import torch
from weaver import ServiceClient, types

# Set your API key
# Create the service client with context manager
with ServiceClient(
    api_key=os.getenv("WEAVER_API_KEY"),
) as service_client:
    # Create a training client for Qwen3-8B
    training_client = service_client.create_model(
        base_model="Qwen/Qwen3-8B"
    )

    # Get the tokenizer
    tokenizer = training_client.get_tokenizer()

    # Prepare training data
    examples = [
        {"input": "hello world", "output": "ello-hay orld-way"},
        {"input": "banana split", "output": "anana-bay plit-say"},
    ]

    # Process examples into training format
    def process_example(example):
        prompt = f"English: {example['input']}\nPig Latin:"
        prompt_tokens = tokenizer.encode(prompt, add_special_tokens=True)
        completion_tokens = tokenizer.encode(f" {example['output']}\n\n", add_special_tokens=False)

        tokens = prompt_tokens + completion_tokens
        weights = [0.0] * len(prompt_tokens) + [1.0] * len(completion_tokens)

        input_tokens = tokens[:-1]
        target_tokens = tokens[1:]
        weights = weights[1:]

        return types.Datum(
            model_input=types.ModelInput.from_ints(input_tokens),
            loss_fn_inputs={
                "target_tokens": torch.tensor(target_tokens, dtype=torch.int64),
                "weights": torch.tensor(weights, dtype=torch.float32),
            },
        )

    processed_examples = [process_example(ex) for ex in examples]

    # Training loop
    adam_params = types.AdamParams(learning_rate=1e-4)
    for step in range(10):
        # Forward and backward pass
        training_client.forward_backward(
            processed_examples,
            "cross_entropy",
            wait=True,
        )

        # Optimizer step
        training_client.optim_step(adam_params, wait=True)
        print(f"Step {step} completed")

    print("Training complete!")

Step 2: Run Training

python train.py

That's it! Weaver will train your model on distributed GPU infrastructure.

Sampling from Your Trained Model

After training, you can sample from your model:

# Save weights for sampling
sampling_client = training_client.save_weights_and_get_sampling_client(
    name="my-model"
)

# Sample from the model
prompt_tokens = tokenizer.encode("English: coffee break\nPig Latin:", add_special_tokens=True)
prompt = types.ModelInput.from_ints(prompt_tokens)

params = types.SamplingParams(max_tokens=20, temperature=0.0, stop=["\n"])
result = sampling_client.sample(
    prompt=prompt,
    sampling_params=params,
    num_samples=1,
)

# Decode and print the response
response_tokens = result["sequences"][0]["tokens"]
response = tokenizer.decode(response_tokens)
print(f"Response: {response}")

Next Steps

Now that you have Weaver installed:

• Learn about Training and Sampling - Core training APIs
• Explore Loss Functions - Available loss functions
• Check out Model Lineup - Supported models
• Read about Saving and Loading - Model persistence

Troubleshooting

Common Issues

Issue: ImportError: No module named 'weaver'

Solution: Make sure you installed the package:

pip install nex-weaver

Issue: Authentication errors

Solution: Verify your API key is set:

echo $WEAVER_API_KEY

Issue: Connection errors

Solution: Check your network connection and API endpoint availability.

For more help, check the specific API documentation or contact support.