Why 1023 in this code?

[ty5491003]

gpt-2-simple/gpt_2_simple/gpt_2.py

Line 435 in e6afb28

length=min(length, 1023 - (len(context_tokens) if prefix else 0)),

From this code, we can see that the maxlen of the generative data is 1023 tokens, no matter how we adjust the '--length' parameter when generating. So i want to ask why set a constant 1023 in this code?

[minimaxir]

The original GPT-2 model is fixed at a context window of 1024; if you go over, it'll error.

The fix is a sliding windows approach which is in #87 but it needs testing.

[ty5491003]

Got it, thx.