mirror of
https://github.com/LostRuins/koboldcpp.git
synced 2026-05-06 08:01:27 +00:00
3f42ed1af7
* fix corner case in sd_oai_transform_params Also fix typo in the function name. * support for customizing loaded LoRA multipliers The `sdloramult` flag now accepts a list of multipliers, one for each LoRA. If all multipliers are non-zero, LoRAs load as before, with no extra VRAM usage or performance impact. If any LoRA has a multiplier of 0, we switch to `at_runtime` mode, and these LoRAs will be available to multiplier changes via the `lora` sdapi field and show up in the `sdapi/v1/loras` endpoint. All LoRAs are still preloaded on startup, and cached to avoid file reloads. If the list of multipliers is shorter than the list of LoRAs, the multiplier list is extended with the first multiplier (1.0 by default), to keep it compatible with the previous behavior. * support for `<lora:name:multiplier>` prompt syntax and metadata * add a few tests for sanitize_lora_multipliers
82 lines
2.3 KiB
Python
82 lines
2.3 KiB
Python
import sys
|
|
import os
|
|
|
|
parent_dir = os.path.abspath(os.path.join(__file__, "..", ".."))
|
|
sys.path.append(parent_dir)
|
|
|
|
import koboldcpp
|
|
|
|
def extract_loras_from_prompt(*args, **kwargs):
|
|
"""
|
|
>>> prompt = "no <lora: tag, even though with a : and 0> it could look like it"
|
|
>>> clean, data = extract_loras_from_prompt(prompt)
|
|
>>> clean
|
|
'no <lora: tag, even though with a : and 0> it could look like it'
|
|
>>> data
|
|
[]
|
|
|
|
>>> prompt = "even after a <lora:valid:1> tag, an unending <lora: tag should be ignored"
|
|
>>> clean, data = extract_loras_from_prompt(prompt)
|
|
>>> clean
|
|
'even after a tag, an unending <lora: tag should be ignored'
|
|
>>> data
|
|
[{'name': 'valid', 'multiplier': 1.0}]
|
|
|
|
>>> prompt = "A portrait <lora:models/face:0.8> with soft lighting"
|
|
>>> clean, data = extract_loras_from_prompt(prompt)
|
|
>>> clean
|
|
'A portrait with soft lighting'
|
|
>>> data
|
|
[{'name': 'models/face', 'multiplier': 0.8}]
|
|
|
|
>>> prompt = "<lora:foo:1.0> start <lora:|high_noise|bar:0.5> end"
|
|
>>> clean, data = extract_loras_from_prompt(prompt)
|
|
>>> clean
|
|
' start end'
|
|
>>> data
|
|
[{'name': 'foo', 'multiplier': 1.0}, {'name': 'bar', 'multiplier': 0.5, 'is_high_noise': True}]
|
|
|
|
>>> prompt = "bad <lora:bad:abc> good <lora:good:2>"
|
|
>>> clean, data = extract_loras_from_prompt(prompt)
|
|
>>> clean
|
|
'bad <lora:bad:abc> good '
|
|
>>> data
|
|
[{'name': 'good', 'multiplier': 2.0}]
|
|
|
|
>>> prompt = "x<lora:a:0.15>y<lora:b:0.2>z"
|
|
>>> clean, data = extract_loras_from_prompt(prompt)
|
|
>>> clean
|
|
'xyz'
|
|
>>> data
|
|
[{'name': 'a', 'multiplier': 0.15}, {'name': 'b', 'multiplier': 0.2}]
|
|
"""
|
|
|
|
return koboldcpp.extract_loras_from_prompt(*args, **kwargs)
|
|
|
|
def sanitize_lora_multipliers(*args, **kwargs):
|
|
"""
|
|
>>> sanitize_lora_multipliers(None)
|
|
[1.0]
|
|
|
|
>>> sanitize_lora_multipliers(0.75)
|
|
[0.75]
|
|
>>> sanitize_lora_multipliers("2")
|
|
[2.0]
|
|
|
|
>>> sanitize_lora_multipliers([0.5, "1.2", 3])
|
|
[0.5, 1.2, 3.0]
|
|
|
|
>>> sanitize_lora_multipliers([])
|
|
[]
|
|
|
|
>>> sanitize_lora_multipliers(["bad", None, ""])
|
|
[0.0, 0.0, 0.0]
|
|
"""
|
|
return koboldcpp.sanitize_lora_multipliers(*args, **kwargs)
|
|
|
|
if __name__ == '__main__':
|
|
import doctest
|
|
failures, _ = doctest.testmod()
|
|
if failures:
|
|
raise SystemExit(f"{failures} doctest{'s' if failures != 1 else ''} failed")
|
|
|