You can deploy this model and start earning money today!

zaballalkala/
astro-1-nano

Name: astro-1-nano
Availability: InStock

Views

No views yet

Astro 1 Nano 🚀

Astro is an advanced 5B parameter artificial intelligence model developed by AstroAI, optimized for high-speed natural conversations, structural text tasks, and local deployment.

Key Features

Efficient Size (5B): High performance with low VRAM footprint, perfect for local inference.
Structured Chat Format: Ready out-of-the-box for system and user instruction following.

🛠️ Quantization (GGUF & Local Run)

Coming soon! We are exporting GGUF versions optimized for Ollama and LM Studio. Stay tuned!

💻 Quick Start & Usage

You can easily run Astro-1-Nano using the Hugging Face transformers library. Ensure you have the chat_template correctly configured.

python

import torch
import json
from huggingface_hub import hf_hub_download
from transformers import AutoModelForCausalLM, AutoTokenizer, Qwen2Config

model_id = "zaballalkala/astro-1-nano"


config_file_path = hf_hub_download(repo_id=model_id, filename="config.json")

with open(config_file_path, "r") as f:
    config_dict = json.load(f)

config_dict["model_type"] = "qwen2"
config_dict["architectures"] = ["Qwen2ForCausalLM"]

config = Qwen2Config.from_dict(config_dict)

tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    config=config,
    torch_dtype=torch.bfloat16,
    device_map="auto",
    trust_remote_code=True
)

messages = [
    {"role": "user", "content": "hi."}
]

text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True
)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)

print("thinking... 🧠\n")
generated_ids = model.generate(
    **model_inputs,
    max_new_tokens=512,
    temperature=0.3,
    do_sample=True
)

generated_ids = [
    output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
]
response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]

print(response)