Public/automatic
1
0
Fork 0
mirror of https://github.com/vladmandic/automatic synced 2026-09-10 06:48:43 +02:00
Code Issues Packages Projects Releases Wiki Activity

add 4bit t5

Browse Source
This commit is contained in:
Vladimir Mandic
2024-06-17 09:39:30 -04:00
parent a1f53add94
commit bbaa03a466
6 changed files with 29 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files
CHANGELOG.md
+3
View File
@@ -13,12 +13,15 @@
- enable taesd preview and non-full quality mode
- enable base LoRA support
- add support for 4bit quantized t5 text encoder
simply select in *settings -> model -> text encoder*
- simplified loading of model in single-file safetensors format
loading sd3 can now be performed fully offline
- add support for nncf compressed weights, thanks @Disty0!
- add support for sampler shift for Euler FlowMatch
see *settings -> samplers*, also available as param in xyz grid
higher shift means model will spend more time on structure and less on details
- add support for selecting text encoder in xyz grid
### Improvements: General
modules/model_sd3.py
+13
View File
@@ -86,6 +86,19 @@ def load_t5(pipe, module, te3=None, cache_dir=None):
torch_dtype=pipe.text_encoder.dtype,
)
setattr(pipe, module, t5)
elif 'fp4' in te3.lower():
modelloader.hf_login()
from installer import install
install('bitsandbytes', quiet=True)
quantization_config = transformers.BitsAndBytesConfig(load_in_4bit=True)
t5 = transformers.T5EncoderModel.from_pretrained(
repo_id,
subfolder='text_encoder_3',
quantization_config=quantization_config,
cache_dir=cache_dir,
torch_dtype=pipe.text_encoder.dtype,
)
setattr(pipe, module, t5)
elif 'fp8' in te3.lower():
modelloader.hf_login()
from installer import install
modules/sd_models.py
+5 -5
View File
@@ -1528,18 +1528,18 @@ def load_model(checkpoint_info=None, already_loaded_state_dict=None, timer=None,
def reload_text_encoder(initial=False):
if initial and (shared.opts.sd_te3 is None or shared.opts.sd_te3 == 'None'):
if initial and (shared.opts.sd_text_encoder is None or shared.opts.sd_text_encoder == 'None'):
return # dont unload
signature = inspect.signature(shared.sd_model.__class__.__init__, follow_wrapped=True, eval_str=True).parameters
t5 = [k for k, v in signature.items() if 'T5EncoderModel' in str(v)]
if len(t5) > 0:
from modules.model_sd3 import load_t5
shared.log.debug(f'Load: t5={shared.opts.sd_te3} module="{t5[0]}"')
load_t5(pipe=shared.sd_model, module=t5[0], te3=shared.opts.sd_te3, cache_dir=shared.opts.diffusers_dir)
shared.log.debug(f'Load: t5={shared.opts.sd_text_encoder} module="{t5[0]}"')
load_t5(pipe=shared.sd_model, module=t5[0], te3=shared.opts.sd_text_encoder, cache_dir=shared.opts.diffusers_dir)
elif hasattr(shared.sd_model, 'text_encoder_3'):
from modules.model_sd3 import load_t5
shared.log.debug(f'Load: t5={shared.opts.sd_te3} module="text_encoder_3"')
load_t5(pipe=shared.sd_model, module='text_encoder_3', te3=shared.opts.sd_te3, cache_dir=shared.opts.diffusers_dir)
shared.log.debug(f'Load: t5={shared.opts.sd_text_encoder} module="text_encoder_3"')
load_t5(pipe=shared.sd_model, module='text_encoder_3', te3=shared.opts.sd_text_encoder, cache_dir=shared.opts.diffusers_dir)
def reload_model_weights(sd_model=None, info=None, reuse_dict=False, op='model', force=False):
modules/shared.py
+1 -1
View File
@@ -391,7 +391,7 @@ options_templates.update(options_section(('sd', "Execution & Models"), {
"sd_model_refiner": OptionInfo('None', "Refiner model", gr.Dropdown, lambda: {"choices": ['None'] + list_checkpoint_tiles()}, refresh=refresh_checkpoints),
"sd_vae": OptionInfo("Automatic", "VAE model", gr.Dropdown, lambda: {"choices": shared_items.sd_vae_items()}, refresh=shared_items.refresh_vae_list),
"sd_unet": OptionInfo("None", "UNET model", gr.Dropdown, lambda: {"choices": shared_items.sd_unet_items()}, refresh=shared_items.refresh_unet_list),
"sd_te3": OptionInfo('None', "Text encoder model", gr.Dropdown, lambda: {"choices": ['None', 'T5 FP8', 'T5 FP16']}),
"sd_text_encoder": OptionInfo('None', "Text encoder model", gr.Dropdown, lambda: {"choices": ['None', 'T5 FP4', 'T5 FP8', 'T5 FP16']}),
"sd_checkpoint_autoload": OptionInfo(True, "Model autoload on start"),
"sd_model_dict": OptionInfo('None', "Use separate base dict", gr.Dropdown, lambda: {"choices": ['None'] + list_checkpoint_tiles()}, refresh=refresh_checkpoints),
"stream_load": OptionInfo(False, "Load models using stream loading method", gr.Checkbox, {"visible": not native }),
scripts/xyz_grid.py
+6
View File
@@ -138,6 +138,11 @@ def apply_vae(p, x, xs):
sd_vae.reload_vae_weights(shared.sd_model, vae_file=find_vae(x))
def apply_te(p, x, xs):
shared.opts.data["sd_text_encoder"] = x
sd_models.reload_text_encoder()
def apply_styles(p: processing.StableDiffusionProcessingTxt2Img, x: str, _):
p.styles.extend(x.split(','))
@@ -230,6 +235,7 @@ axis_options = [
AxisOption("Prompt S/R", str, apply_prompt, fmt=format_value),
AxisOption("Model", str, apply_checkpoint, fmt=format_value, cost=1.0, choices=lambda: sorted(sd_models.checkpoints_list)),
AxisOption("VAE", str, apply_vae, cost=0.7, choices=lambda: ['None'] + list(sd_vae.vae_dict)),
AxisOption("Text encoder", str, apply_te, cost=0.7, choices=lambda: ['None', 'T5 FP4', 'T5 FP8', 'T5 FP16']),
AxisOption("Styles", str, apply_styles, choices=lambda: [s.name for s in shared.prompt_styles.styles.values()]),
AxisOption("Seed", int, apply_field("seed")),
AxisOption("Steps", int, apply_field("steps")),
webui.py
+1 -1
View File
@@ -168,7 +168,7 @@ def load_model():
thread_refiner.join()
shared.opts.onchange("sd_model_checkpoint", wrap_queued_call(lambda: modules.sd_models.reload_model_weights(op='model')), call=False)
shared.opts.onchange("sd_model_refiner", wrap_queued_call(lambda: modules.sd_models.reload_model_weights(op='refiner')), call=False)
shared.opts.onchange("sd_te3", wrap_queued_call(lambda: modules.sd_models.reload_text_encoder()), call=False)
shared.opts.onchange("sd_text_encoder", wrap_queued_call(lambda: modules.sd_models.reload_text_encoder()), call=False)
shared.opts.onchange("sd_model_dict", wrap_queued_call(lambda: modules.sd_models.reload_model_weights(op='dict')), call=False)
shared.opts.onchange("sd_vae", wrap_queued_call(lambda: modules.sd_vae.reload_vae_weights()), call=False)
shared.opts.onchange("sd_backend", wrap_queued_call(lambda: modules.sd_models.change_backend()), call=False)