vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

reduce max ctx to fit instead of crashing

Browse source
This commit is contained in:
Concedo 2023-11-27 19:08:54 +08:00
parent 8acd7be734
commit 0e5f16de53
1 changed files with 5 additions and 3 deletions
Download patch file Download diff file Expand all files Collapse all files

8
koboldcpp.py
View file

@ -310,9 +310,11 @@ def generate(prompt, memory="", max_length=32, max_context_length=512, temperatu
max_length = max_context_length-1 max_length = max_context_length-1
inputs.max_context_length = max_context_length # this will resize the context buffer if changed inputs.max_context_length = max_context_length # this will resize the context buffer if changed
global showmaxctxwarning global showmaxctxwarning
if showmaxctxwarning and max_context_length > maxctx: if max_context_length > maxctx:
print(f"\n(Warning! Request max_context_length={max_context_length} exceeds allocated context size of {maxctx}. Consider launching with increased --contextsize to avoid errors. This message will only show once per session.)") if showmaxctxwarning:
showmaxctxwarning = False print(f"\n(Warning! Request max_context_length={max_context_length} exceeds allocated context size of {maxctx}. It will be reduced to fit. Consider launching with increased --contextsize to avoid errors. This message will only show once per session.)")
showmaxctxwarning = False
max_context_length = maxctx
inputs.max_length = max_length inputs.max_length = max_length
inputs.temperature = temperature inputs.temperature = temperature
inputs.top_k = top_k inputs.top_k = top_k