from transformers import AutoTokenizer, AutoModelForCausalLM
from optimum.quanto import QuantizedModelForCausalLM

# Load tokenizer and model directly
tokenizer = AutoTokenizer.from_pretrained("CarlOwOs/Qwen3-0.6B-Base-int2", trust_remote_code=True)
model = QuantizedModelForCausalLM.from_pretrained("CarlOwOs/Qwen3-0.6B-Base-int2")

# Generate text
inputs = tokenizer("Hello, how are you?", return_tensors="pt")
outputs = model.generate(**inputs, max_length=50, do_sample=True, temperature=0.7)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Alternative Loading Method

python

# If the direct method doesn't work, try this:
from transformers import AutoTokenizer
from optimum.quanto import QuantizedModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("CarlOwOs/Qwen3-0.6B-Base-int2", trust_remote_code=True)
model = QuantizedModelForCausalLM.from_pretrained("CarlOwOs/Qwen3-0.6B-Base-int2")

# Use the model for inference
inputs = tokenizer("What is the capital of France?", return_tensors="pt")
with torch.no_grad():
    outputs = model.generate(
        **inputs, 
        max_length=100, 
        do_sample=True, 
        temperature=0.7,
        pad_token_id=tokenizer.eos_token_id
    )
generated_text = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(generated_text)

Performance

This quantized model provides significant memory savings compared to the original model:

Inference Speed: Similar to original model
Quality: Maintains good performance for most tasks

Technical Details

Quantization Method: optimum-quanto int2 weight quantization
Base Model: Qwen/Qwen3-0.6B-Base
Precision: int2 weights, float16 activations

License

Same as the base model license.