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

reduce number of retries, fixed maxlength > maxctx bug

Browse source
This commit is contained in:
Concedo 2023-07-23 09:59:34 +08:00
parent fa0270df7c
commit 0cca0726fe
1 changed files with 3 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

4
koboldcpp.py
View file

@ -222,6 +222,8 @@ def generate(prompt,max_length=20, max_context_length=512, temperature=0.8, top_
inputs = generation_inputs() inputs = generation_inputs()
outputs = ctypes.create_unicode_buffer(ctypes.sizeof(generation_outputs)) outputs = ctypes.create_unicode_buffer(ctypes.sizeof(generation_outputs))
inputs.prompt = prompt.encode("UTF-8") inputs.prompt = prompt.encode("UTF-8")
if max_length >= max_context_length:
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
inputs.max_length = max_length inputs.max_length = max_length
inputs.temperature = temperature inputs.temperature = temperature
@ -1426,7 +1428,7 @@ def run_horde_worker(args, api_key, worker_name):
break break
else: else:
currentjob_attempts += 1 currentjob_attempts += 1
if currentjob_attempts>10: if currentjob_attempts>5:
break break
print("Server Busy - Not ready to generate...") print("Server Busy - Not ready to generate...")
time.sleep(5) time.sleep(5)