amanrs

2 lata temudiscuss

This is harder than it looks.

First "token-healing" doesn't work. Consider the case "app" where the most likely options are "ap|praisal" or "apple|sauce". You can't just sample all tokens that start with app, or you'd miss appraisal.

Second, it's easy to come up with a naive algorithm that samples from the true distribution. It's very difficult to make this algorithm efficient.

amanrs

2 lata temudiscuss

It edits the file based on the "plan" laid out by a smarter language model

amanrs

3 lata temudiscuss

The key misconception about many quantization methods is that lower precision = better speed.

I believe GPT-Q is not much faster than bf16 from skimming the AWQ paper - https://arxiv.org/pdf/2306.00978.pdf

It's 3x faster for a batch size of 1, but that's still over 10x more expensive than gpt-3.5

For larger batch sizes, bf16 costs dip below 3-bit quantized.

Submissions

Editing Files at 1000 tokens/s with llama-70B

comments