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

fixed length exceeding max ctx

Browse source
This commit is contained in:
Concedo 2023-12-14 16:58:41 +08:00
parent 04bd895311
commit f0de4953ae
1 changed files with 1 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

2
koboldcpp.py
View file

@ -310,13 +310,13 @@ def generate(prompt, memory="", max_length=32, max_context_length=512, temperatu
inputs.memory = memory.encode("UTF-8") inputs.memory = memory.encode("UTF-8")
if max_length >= max_context_length: if max_length >= max_context_length:
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
global showmaxctxwarning global showmaxctxwarning
if max_context_length > maxctx: if max_context_length > maxctx:
if showmaxctxwarning: if showmaxctxwarning:
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.)") 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 showmaxctxwarning = False
max_context_length = maxctx max_context_length = maxctx
inputs.max_context_length = max_context_length # this will resize the context buffer if changed
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