vrr/koboldcpp
2
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2025-09-17 04:19:40 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

warning for max tokens being too high

Browse source
This commit is contained in:
Concedo 2025-04-02 18:58:38 +08:00
parent 669311365c
commit 7f1003be44
1 changed files with 8 additions and 5 deletions
Download patch file Download diff file Expand all files Collapse all files

13
koboldcpp.py
View file

@ -1228,13 +1228,16 @@ def generate(genparams, stream_flag=False):
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!!! ====== !!!\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 issues. This message will only show once per session.)\n!!! ====== !!!")
showmaxctxwarning = False showmaxctxwarning = False
max_context_length = maxctx max_context_length = maxctx
min_remain = min(max_context_length-4, 16) min_remain_hardlimit = max(min(max_context_length-4, 16),int(max_context_length*0.1))
if max_length >= (max_context_length-min_remain): min_remain_softlimit = max(min(max_context_length-4, 16),int(max_context_length*0.4))
max_length = max_context_length-min_remain if max_length >= (max_context_length-min_remain_softlimit):
print("\nWarning: You are trying to generate with max_length near or exceeding max_context_length. Most of the context will be removed, and your outputs will not be very coherent.") print(f"\n!!! ====== !!!\nWarning: You are trying to generate text with max_length ({max_length}) near or exceeding max_context_length limit ({max_context_length}).\nMost of the context will be removed, and your outputs will not be very coherent.\nConsider launching with increased --contextsize to avoid issues.\n!!! ====== !!!")
if max_length >= (max_context_length-min_remain_hardlimit):
max_length = max_context_length-min_remain_hardlimit
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
inputs.max_length = max_length inputs.max_length = max_length