Upload app.py
Browse files
app.py
CHANGED
|
@@ -1,4 +1,5 @@
|
|
| 1 |
import os
|
|
|
|
| 2 |
import gradio as gr
|
| 3 |
import torch
|
| 4 |
from TeLVE.imagine import ImageCaptioningModel, load_model, generate_caption
|
|
@@ -9,8 +10,22 @@ from huggingface_hub import hf_hub_download, list_repo_files
|
|
| 9 |
MODELS_DIR = "./TeLVE/models"
|
| 10 |
TOKENIZER_PATH = "./TeLVE/tokenizer"
|
| 11 |
HF_REPO_ID = "outsu/TeLVE"
|
|
|
|
| 12 |
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
| 13 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 14 |
def list_available_models():
|
| 15 |
"""List all .pth models in the models directory"""
|
| 16 |
if not os.path.exists(MODELS_DIR):
|
|
@@ -27,14 +42,19 @@ def get_hf_model_list():
|
|
| 27 |
return []
|
| 28 |
|
| 29 |
def download_missing_models():
|
| 30 |
-
"""Download missing models from HuggingFace"""
|
| 31 |
if not os.path.exists(MODELS_DIR):
|
| 32 |
os.makedirs(MODELS_DIR)
|
| 33 |
|
|
|
|
|
|
|
| 34 |
local_models = set(list_available_models())
|
| 35 |
hf_models = set(get_hf_model_list())
|
| 36 |
|
| 37 |
-
for
|
|
|
|
|
|
|
|
|
|
| 38 |
try:
|
| 39 |
print(f"Downloading missing model: {model}")
|
| 40 |
hf_hub_download(
|
|
@@ -43,8 +63,21 @@ def download_missing_models():
|
|
| 43 |
local_dir=os.path.dirname(MODELS_DIR),
|
| 44 |
local_dir_use_symlinks=False
|
| 45 |
)
|
|
|
|
| 46 |
except Exception as e:
|
| 47 |
print(f"Error downloading {model}: {str(e)}")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 48 |
|
| 49 |
def generate_description(image, model_name):
|
| 50 |
"""Generate image caption using selected model"""
|
|
@@ -96,6 +129,8 @@ def create_interface():
|
|
| 96 |
return interface
|
| 97 |
|
| 98 |
if __name__ == "__main__":
|
|
|
|
|
|
|
| 99 |
print("Checking for missing models...")
|
| 100 |
download_missing_models()
|
| 101 |
demo = create_interface()
|
|
|
|
| 1 |
import os
|
| 2 |
+
import json
|
| 3 |
import gradio as gr
|
| 4 |
import torch
|
| 5 |
from TeLVE.imagine import ImageCaptioningModel, load_model, generate_caption
|
|
|
|
| 10 |
MODELS_DIR = "./TeLVE/models"
|
| 11 |
TOKENIZER_PATH = "./TeLVE/tokenizer"
|
| 12 |
HF_REPO_ID = "outsu/TeLVE"
|
| 13 |
+
MODEL_STATE_FILE = "./TeLVE/model_state.json"
|
| 14 |
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
| 15 |
|
| 16 |
+
def load_model_state():
|
| 17 |
+
"""Load the model state from JSON file"""
|
| 18 |
+
if os.path.exists(MODEL_STATE_FILE):
|
| 19 |
+
with open(MODEL_STATE_FILE, 'r') as f:
|
| 20 |
+
return json.load(f)
|
| 21 |
+
return {"downloaded_models": []}
|
| 22 |
+
|
| 23 |
+
def save_model_state(state):
|
| 24 |
+
"""Save the model state to JSON file"""
|
| 25 |
+
os.makedirs(os.path.dirname(MODEL_STATE_FILE), exist_ok=True)
|
| 26 |
+
with open(MODEL_STATE_FILE, 'w') as f:
|
| 27 |
+
json.dump(state, f)
|
| 28 |
+
|
| 29 |
def list_available_models():
|
| 30 |
"""List all .pth models in the models directory"""
|
| 31 |
if not os.path.exists(MODELS_DIR):
|
|
|
|
| 42 |
return []
|
| 43 |
|
| 44 |
def download_missing_models():
|
| 45 |
+
"""Download missing models from HuggingFace with state management"""
|
| 46 |
if not os.path.exists(MODELS_DIR):
|
| 47 |
os.makedirs(MODELS_DIR)
|
| 48 |
|
| 49 |
+
state = load_model_state()
|
| 50 |
+
downloaded_models = set(state["downloaded_models"])
|
| 51 |
local_models = set(list_available_models())
|
| 52 |
hf_models = set(get_hf_model_list())
|
| 53 |
|
| 54 |
+
# Check for models that need downloading
|
| 55 |
+
models_to_download = (hf_models - local_models) - downloaded_models
|
| 56 |
+
|
| 57 |
+
for model in models_to_download:
|
| 58 |
try:
|
| 59 |
print(f"Downloading missing model: {model}")
|
| 60 |
hf_hub_download(
|
|
|
|
| 63 |
local_dir=os.path.dirname(MODELS_DIR),
|
| 64 |
local_dir_use_symlinks=False
|
| 65 |
)
|
| 66 |
+
downloaded_models.add(model)
|
| 67 |
except Exception as e:
|
| 68 |
print(f"Error downloading {model}: {str(e)}")
|
| 69 |
+
continue
|
| 70 |
+
|
| 71 |
+
# Update state with newly downloaded models
|
| 72 |
+
state["downloaded_models"] = list(downloaded_models)
|
| 73 |
+
save_model_state(state)
|
| 74 |
+
|
| 75 |
+
def verify_model_integrity():
|
| 76 |
+
"""Verify that all models in state actually exist"""
|
| 77 |
+
state = load_model_state()
|
| 78 |
+
local_models = set(list_available_models())
|
| 79 |
+
state["downloaded_models"] = list(set(state["downloaded_models"]) & local_models)
|
| 80 |
+
save_model_state(state)
|
| 81 |
|
| 82 |
def generate_description(image, model_name):
|
| 83 |
"""Generate image caption using selected model"""
|
|
|
|
| 129 |
return interface
|
| 130 |
|
| 131 |
if __name__ == "__main__":
|
| 132 |
+
print("Verifying model integrity...")
|
| 133 |
+
verify_model_integrity()
|
| 134 |
print("Checking for missing models...")
|
| 135 |
download_missing_models()
|
| 136 |
demo = create_interface()
|