GPTQ: Dequantizing Weights
Beginner
The problem, from first principles
Inference with a quantized model multiplies the stored integer codes by a scale to recover approximate weights, usually inside the matrix kernel. Dequantization is that multiplication.
From theory to code
Implement dequantize(q, scale), returning the approximate float weights.
Constraints
- Return floats.
Hints
Hint 1
Multiply the integer codes by the scale.
Keywords
research-paperssystemsquantizationinference
Python
Run your code to see results
No results yet
Run your code to check it against the examples, or submit to run every test.