Public/automatic
1
0
Fork 0
mirror of https://github.com/vladmandic/automatic synced 2026-09-18 16:54:33 +02:00
Code Issues Packages Projects Releases Wiki Activity

OpenVINO fix model reloading

Browse Source
This commit is contained in:
Disty0
2023-08-23 00:31:43 +03:00
parent 6a4d4ea5b7
commit 863fa38c24
4 changed files with 18 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/intel/openvino/__init__.py
+9 -2
View File
@@ -1,6 +1,6 @@
import os
import torch
from openvino.frontend.pytorch.torchdynamo.execute import execute
from openvino.frontend.pytorch.torchdynamo.execute import execute, partitioned_modules, compiled_cache
from openvino.frontend.pytorch.torchdynamo.partition import Partitioner
from openvino.runtime import Core, Type, PartialShape
from torch._dynamo.backends.common import fake_tensor_unsupported
@@ -12,7 +12,7 @@ from hashlib import sha256
class ModelState:
def __init__(self):
self.recompile = 1
self.device = "CPU"
self.device = "GPU"
self.height = 512
self.width = 512
self.batch_size = 1
@@ -112,3 +112,10 @@ def get_cached_file_name(*args, model_hash_str, device, cache_root):
file_name = None
model_hash_str = None
return file_name
def openvino_clear_caches():
global partitioned_modules
global compiled_cache
compiled_cache.clear()
partitioned_modules.clear()
modules/sd_hijack.py
+4 -1
View File
@@ -189,7 +189,10 @@ class StableDiffusionModelHijack:
shared.log.info(f"Compiling pipeline={m.model.__class__.__name__} mode={opts.cuda_compile_backend}")
import torch._dynamo # pylint: disable=unused-import,redefined-outer-name
if shared.opts.cuda_compile_backend == "openvino_fx":
from modules.intel.openvino import openvino_fx
torch._dynamo.reset()
from modules.intel.openvino import openvino_fx, openvino_clear_caches, model_state # pylint: disable=unused-import
openvino_clear_caches()
model_state.partition_id = 0
log_level = logging.WARNING if opts.cuda_compile_verbose else logging.CRITICAL # pylint: disable=protected-access
if hasattr(torch, '_logging'):
torch._logging.set_logs(dynamo=log_level, aot=log_level, inductor=log_level) # pylint: disable=protected-access
modules/sd_models.py
+4 -1
View File
@@ -803,7 +803,10 @@ def load_diffuser(checkpoint_info=None, already_loaded_state_dict=None, timer=No
shared.log.info(f"Compiling pipeline={sd_model.__class__.__name__} shape={8 * sd_model.unet.config.sample_size} mode={shared.opts.cuda_compile_backend}")
import torch._dynamo # pylint: disable=unused-import,redefined-outer-name
if shared.opts.cuda_compile_backend == "openvino_fx":
from modules.intel.openvino import openvino_fx # pylint: disable=unused-import
torch._dynamo.reset()
from modules.intel.openvino import openvino_fx, openvino_clear_caches, model_state # pylint: disable=unused-import
openvino_clear_caches()
model_state.partition_id = 0
log_level = logging.WARNING if shared.opts.cuda_compile_verbose else logging.CRITICAL # pylint: disable=protected-access
if hasattr(torch, '_logging'):
torch._logging.set_logs(dynamo=log_level, aot=log_level, inductor=log_level) # pylint: disable=protected-access
modules/shared.py
+1 -1
View File
@@ -418,7 +418,7 @@ options_templates.update(options_section(('diffusers', "Diffusers Settings"), {
"diffusers_attention_slicing": OptionInfo(False, "Enable attention slicing"),
"diffusers_model_load_variant": OptionInfo("default", "Diffusers model loading variant", gr.Radio, lambda: {"choices": ['default', 'fp32', 'fp16']}),
"diffusers_vae_load_variant": OptionInfo("default", "Diffusers VAE loading variant", gr.Radio, lambda: {"choices": ['default', 'fp32', 'fp16']}),
"diffusers_lora_loader": OptionInfo("sequential apply", "Diffusers LoRA loading variant", gr.Radio, lambda: {"choices": ['sequential apply', 'merge and apply', 'diffusers default']}),
"diffusers_lora_loader": OptionInfo("diffusers default" if cmd_opts.use_openvino else "sequential apply", "Diffusers LoRA loading variant", gr.Radio, lambda: {"choices": ['sequential apply', 'merge and apply', 'diffusers default']}),
# "diffusers_force_zeros": OptionInfo(False, "Force zeros for prompts when empty"),
# "diffusers_aesthetics_score": OptionInfo(6.0, "Require aesthetic score", gr.Slider, {"minimum": 0, "maximum": 10, "step": 0.1}),
}))