Phi-1.5 (1.3B Outperforms Llama 2 7B) (huggingface.co)

3 points by jcuenod 2y ago ↗ HN

3 comments

[ 3.0 ms ] story [ 74.0 ms ] thread

I am little worried about their lack of grounding or hallucination presentation for coding examples

the model size is above 2GB, is that a quantized version? basically want to know if there is a room for further compression.

I didn't see mention of quantization. I think the point is how few parameters were needed for the performance results they were able to achieve.