Mirrored_Repos/InvokeAI
1
0
Fork 0
mirror of https://github.com/invoke-ai/InvokeAI synced 2024-08-30 20:32:17 +00:00
Code Issues Packages Projects Releases Wiki Activity

reduce VRAM cache default; take max RAM from system

Browse Source
This commit is contained in:
Lincoln Stein 2023-08-02 14:27:13 -04:00
parent 29ac252501
commit 5de42be4a6
2 changed files with 11 additions and 12 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
invokeai/app/services/config.py
View File

@ -24,7 +24,7 @@ InvokeAI:
sequential_guidance: false
precision: float16
max_cache_size: 6
max_vram_cache_size: 2.7
max_vram_cache_size: 0.5
always_use_cpu: false
free_gpu_mem: false
Features:
@ -173,8 +173,7 @@ from typing import ClassVar, Dict, List, Set, Literal, Union, get_origin, get_ty
INIT_FILE = Path("invokeai.yaml")
DB_FILE = Path("invokeai.db")
LEGACY_INIT_FILE = Path("invokeai.init")
DEFAULT_MAX_VRAM = 2.75
DEFAULT_MAX_VRAM = 0.5
class InvokeAISettings(BaseSettings):
"""

18
invokeai/backend/install/invokeai_configure.py
View File

@ -10,6 +10,7 @@ import sys
import argparse
import io
import os
import psutil
import shutil
import textwrap
import torch
@ -79,10 +80,13 @@ Default_config_file = config.model_conf_path
SD_Configs = config.legacy_conf_path
PRECISION_CHOICES = ["auto", "float16", "float32"]
GB = 1073741824 # GB in bytes
HAS_CUDA = torch.cuda.is_available()
_, MAX_VRAM = torch.cuda.mem_get_info() if HAS_CUDA else (0, 0)
MAX_VRAM /= 1073741824 # GB in bytes
MAX_VRAM_CACHE_RATIO = 0.55 # first guess of optimal vram cache based on total available
MAX_VRAM /= GB
MAX_RAM = psutil.virtual_memory().total / GB
INIT_FILE_PREAMBLE = """# InvokeAI initialization file
# This is the InvokeAI initialization file, which contains command-line default values.
@ -391,9 +395,9 @@ Use cursor arrows to make a checkbox selection, and space to toggle.
)
self.max_cache_size = self.add_widget_intelligent(
IntTitleSlider,
name="RAM cache size. The larger this is, the more models can be kept in memory rather than loading from disk each time (GB)",
name="RAM cache size (GB). Make this at least large enough to hold a single model. Larger sizes will allow you to switch between models quickly without reading from disk.",
value=old_opts.max_cache_size,
out_of=20,
out_of=MAX_RAM,
lowest=3,
begin_entry_at=6,
scroll_exit=True,
@ -402,7 +406,7 @@ Use cursor arrows to make a checkbox selection, and space to toggle.
self.nextrely += 1
self.add_widget_intelligent(
npyscreen.TitleFixedText,
name="VRAM cache size. Make this large enough to hold an entire model, but not more than half your available VRAM (GB)",
name="VRAM cache size (GB). Reserving a small amount of VRAM will modestly speed up the start of image generation.",
begin_entry_at=0,
editable=False,
color="CONTROL",
@ -416,7 +420,6 @@ Use cursor arrows to make a checkbox selection, and space to toggle.
lowest=0.0,
relx=8,
step=0.25,
begin_entry_at=MAX_VRAM * 0.55,
scroll_exit=True,
)
else:
@ -569,9 +572,6 @@ def edit_opts(program_opts: Namespace, invokeai_opts: Namespace) -> argparse.Nam
def default_startup_options(init_file: Path) -> Namespace:
opts = InvokeAIAppConfig.get_config()
# dynamically adust vram for memory size
if not init_file.exists():
opts.max_vram_cache_size = round((MAX_VRAM * MAX_VRAM_CACHE_RATIO) * 4) / 4
return opts