Skip to content

C
coderai
Commit 9af89755 authored by Your Name's avatar Your Name
Browse files
Options

Improve GGUF image model loading - better URL handling 

- Check if model is URL before any processing
- Use original model name with query params for URL download
- Strip query params only for HuggingFace repo ID parsing
- Added more debug output to trace issues
parent 6bc9af36
Hide whitespace changes
Inline Side-by-side
Showing with 98 additions and 15 deletions
coderai
View file @ 9af89755
...@@ -4399,43 +4399,56 @@ def main(): ...@@ -4399,43 +4399,56 @@ def main():
if image_models: if image_models:
print(f"Pre-loading image model: {image_models[0]}") print(f"Pre-loading image model: {image_models[0]}")
# Check if the image model is a GGUF model # Get the original model name
model_name = image_models[0] original_model_name = image_models[0]
# Check if it's a URL first (before any processing)
is_url = original_model_name.startswith('http://') or original_model_name.startswith('https://')
# Strip query parameters from URL if present # Strip query parameters from URL if present
model_name = original_model_name
if '?' in model_name: if '?' in model_name:
model_name = model_name.split('?')[0] model_name = model_name.split('?')[0]
# Check if the image model is a GGUF model
is_gguf = model_name.endswith('.gguf') or 'gguf' in model_name.lower() is_gguf = model_name.endswith('.gguf') or 'gguf' in model_name.lower()
if is_gguf: if is_gguf:
# Load GGUF image model using llama.cpp (VulkanBackend) # Load GGUF image model using llama.cpp
print(f"Detected GGUF image model, loading with llama.cpp...") print(f"Detected GGUF image model, loading with llama.cpp...")
try: try:
from llama_cpp import Llama from llama_cpp import Llama
import os import os
model_key = f"image:{model_name}" model_key = f"image:{original_model_name}"
# Download GGUF model if needed (similar to VulkanBackend) # Download GGUF model if needed (similar to VulkanBackend)
model_path = None model_path = None
if model_name.startswith('http://') or model_name.startswith('https://'):
cached_path = get_cached_model_path(model_name) # Check if it's a URL - download directly
if is_url:
print(f"Image model is a URL: {original_model_name}")
cached_path = get_cached_model_path(original_model_name)
if cached_path: if cached_path:
model_path = cached_path model_path = cached_path
print(f"Using cached GGUF model: {model_path}") print(f"Using cached GGUF model: {model_path}")
else: else:
print(f"Downloading GGUF model: {model_name}") print(f"Downloading GGUF model: {original_model_name}")
cache_dir = get_model_cache_dir() cache_dir = get_model_cache_dir()
model_path = download_model(model_name, cache_dir) model_path = download_model(original_model_name, cache_dir)
elif os.path.isfile(model_name): elif os.path.isfile(model_name):
# Local file
model_path = model_name model_path = model_name
print(f"Loading local GGUF model: {model_path}") print(f"Loading local GGUF model: {model_path}")
else: else:
# Try to download from HuggingFace Hub # Try to download from HuggingFace Hub
print(f"Trying to resolve as HuggingFace model: {model_name}")
try: try:
from huggingface_hub import hf_hub_download, list_repo_files from huggingface_hub import hf_hub_download, list_repo_files
parts = model_name.split('/') parts = model_name.split('/')
if len(parts) >= 2: if len(parts) >= 2:
repo_id = f"{parts[0]}/{parts[1]}" repo_id = f"{parts[0]}/{parts[1]}"
print(f"Looking for GGUF files in repo: {repo_id}")
files = list_repo_files(repo_id) files = list_repo_files(repo_id)
gguf_files = [f for f in files if f.endswith('.gguf')] gguf_files = [f for f in files if f.endswith('.gguf')]
if not gguf_files: if not gguf_files:
...@@ -4449,12 +4462,11 @@ def main(): ...@@ -4449,12 +4462,11 @@ def main():
model_path = None model_path = None
if model_path and os.path.isfile(model_path): if model_path and os.path.isfile(model_path):
# Load with llama.cpp - use Vulkan backend for GGUF # Load with llama.cpp
# GGUF models for image generation need special handling
# Most llama.cpp based image models need GPU layers
n_gpu_layers = -1 # Load all layers to GPU n_gpu_layers = -1 # Load all layers to GPU
n_ctx = 2048 n_ctx = 2048
print(f"Loading GGUF model from: {model_path}")
llama_model = Llama( llama_model = Llama(
model_path=model_path, model_path=model_path,
n_gpu_layers=n_gpu_layers, n_gpu_layers=n_gpu_layers,
...@@ -4462,7 +4474,7 @@ def main(): ...@@ -4462,7 +4474,7 @@ def main():
verbose=False, verbose=False,
) )
multi_model_manager.add_model(model_key, llama_model) multi_model_manager.add_model(model_key, llama_model)
print(f"GGUF image model loaded successfully: {model_name}") print(f"GGUF image model loaded successfully: {original_model_name}")
else: else:
print(f"Could not load GGUF image model: no valid model path") print(f"Could not load GGUF image model: no valid model path")
...@@ -4771,20 +4783,91 @@ def main(): ...@@ -4771,20 +4783,91 @@ def main():
# Pre-load image model if it's the only model configured # Pre-load image model if it's the only model configured
if not model_names and not audio_models and not args.tts_model: if not model_names and not audio_models and not args.tts_model:
print(f"Pre-loading image model...") print(f"Pre-loading image model...")
# Check if the image model is a GGUF model
model_name = image_models[0] # Get the original model name
original_model_name = image_models[0]
# Check if it's a URL first (before any processing)
is_url = original_model_name.startswith('http://') or original_model_name.startswith('https://')
# Strip query parameters from URL if present # Strip query parameters from URL if present
model_name = original_model_name
if '?' in model_name: if '?' in model_name:
model_name = model_name.split('?')[0] model_name = model_name.split('?')[0]
# Check if the image model is a GGUF model
is_gguf = model_name.endswith('.gguf') or 'gguf' in model_name.lower() is_gguf = model_name.endswith('.gguf') or 'gguf' in model_name.lower()
if is_gguf: if is_gguf:
# Load GGUF image model using llama.cpp (VulkanBackend) # Load GGUF image model using llama.cpp
print(f"Detected GGUF image model, loading with llama.cpp...") print(f"Detected GGUF image model, loading with llama.cpp...")
try: try:
from llama_cpp import Llama from llama_cpp import Llama
import os import os
model_key = f"image:{original_model_name}"
# Download GGUF model if needed
model_path = None
# Check if it's a URL - download directly
if is_url:
print(f"Image model is a URL: {original_model_name}")
cached_path = get_cached_model_path(original_model_name)
if cached_path:
model_path = cached_path
print(f"Using cached GGUF model: {model_path}")
else:
print(f"Downloading GGUF model: {original_model_name}")
cache_dir = get_model_cache_dir()
model_path = download_model(original_model_name, cache_dir)
elif os.path.isfile(model_name):
model_path = model_name
print(f"Loading local GGUF model: {model_name}")
else:
# Try to download from HuggingFace Hub
print(f"Trying to resolve as HuggingFace model: {model_name}")
try:
from huggingface_hub import hf_hub_download, list_repo_files
parts = model_name.split('/')
if len(parts) >= 2:
repo_id = f"{parts[0]}/{parts[1]}"
print(f"Looking for GGUF files in repo: {repo_id}")
files = list_repo_files(repo_id)
gguf_files = [f for f in files if f.endswith('.gguf')]
if not gguf_files:
raise ValueError(f"No GGUF files found in {repo_id}")
filename = gguf_files[0]
model_path = hf_hub_download(repo_id=repo_id, filename=filename)
print(f"Downloaded GGUF model to: {model_path}")
except Exception as e:
print(f"Could not resolve GGUF model path: {e}")
model_path = None
if model_path and os.path.isfile(model_path):
n_gpu_layers = -1
n_ctx = 2048
print(f"Loading GGUF model from: {model_path}")
llama_model = Llama(
model_path=model_path,
n_gpu_layers=n_gpu_layers,
n_ctx=n_ctx,
verbose=False,
)
multi_model_manager.add_model(model_key, llama_model)
print(f"GGUF image model loaded successfully: {original_model_name}")
else:
print(f"Could not load GGUF image model: no valid model path")
except ImportError as e:
print(f"Warning: llama_cpp not installed: {e}")
except Exception as e:
print(f"Warning: Failed to pre-load GGUF image model: {e}")
try:
from llama_cpp import Llama
import os
model_key = f"image:{model_name}" model_key = f"image:{model_name}"
# Download GGUF model if needed (similar to VulkanBackend) # Download GGUF model if needed (similar to VulkanBackend)
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment