Update main.py
Browse files
main.py
CHANGED
@@ -5,12 +5,12 @@ from fastapi import FastAPI, File, UploadFile, Form, HTTPException
|
|
5 |
# from fastapi.templating import Jinja2Templates
|
6 |
# from fastapi.responses import FileResponse
|
7 |
|
8 |
-
# Removed 'requests'
|
9 |
# import requests
|
10 |
-
import base64 # Keep if
|
11 |
import os
|
12 |
import random
|
13 |
-
from typing import IO
|
14 |
|
15 |
# Import necessary classes from transformers
|
16 |
import torch
|
@@ -19,6 +19,11 @@ from transformers import AutoModelForCausalLM, AutoTokenizer
|
|
19 |
# Import the Gradio Client and handle_file
|
20 |
from gradio_client import Client, handle_file
|
21 |
|
|
|
|
|
|
|
|
|
|
|
22 |
from deep_translator import GoogleTranslator
|
23 |
from deep_translator.exceptions import InvalidSourceOrTargetLanguage
|
24 |
|
@@ -26,10 +31,7 @@ from deep_translator.exceptions import InvalidSourceOrTargetLanguage
|
|
26 |
app = FastAPI()
|
27 |
|
28 |
# --- Hugging Face Language Model Setup (Local Inference) ---
|
29 |
-
#
|
30 |
-
# Chosen for balance of quality and speed on CPU basic (faster than Gemma 2B, better than GPT-2 base)
|
31 |
-
# If you get access to Gemma 2B-IT and prefer its quality (accepting slower speed),
|
32 |
-
# change this to "google/gemma-2b-it" and use the generate_story_gemma function below.
|
33 |
model_name = "TinyLlama/TinyLlama-1.1B-Chat-v1.0"
|
34 |
tokenizer = None
|
35 |
model = None
|
@@ -47,23 +49,17 @@ def load_language_model():
|
|
47 |
print(f"Loading language model: {model_name}...")
|
48 |
try:
|
49 |
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
50 |
-
# TinyLlama might not have an explicit pad token, setting it to EOS is common practice
|
51 |
if tokenizer.pad_token is None:
|
52 |
tokenizer.pad_token = tokenizer.eos_token
|
53 |
|
54 |
-
# Load model weights. Using float16 to reduce memory footprint on CPU.
|
55 |
model = AutoModelForCausalLM.from_pretrained(
|
56 |
model_name,
|
57 |
torch_dtype=torch.float16, # Use float16 precision
|
58 |
-
# device_map="auto" # Not needed for single CPU
|
59 |
)
|
60 |
-
# model.to("cpu") # Explicitly ensure it's on CPU, although from_pretrained does this by default on CPU-only systems.
|
61 |
|
62 |
print(f"Language model {model_name} loaded successfully.")
|
63 |
except Exception as e:
|
64 |
print(f"Error loading language model {model_name}: {e}")
|
65 |
-
# Depending on requirements, you might want to exit or set model/tokenizer to None permanently
|
66 |
-
# For now, we let the app start, but subsequent generation calls will fail gracefully.
|
67 |
tokenizer = None
|
68 |
model = None
|
69 |
|
@@ -73,69 +69,85 @@ def initialize_caption_client():
|
|
73 |
global caption_client
|
74 |
print(f"Initializing Gradio client for {CAPTION_SPACE_URL}...")
|
75 |
try:
|
76 |
-
# If the target
|
77 |
-
# uncomment
|
78 |
# HF_TOKEN = os.environ.get("HF_TOKEN")
|
79 |
# if HF_TOKEN:
|
80 |
# caption_client = Client(CAPTION_SPACE_URL, hf_token=HF_TOKEN)
|
81 |
# else:
|
82 |
# caption_client = Client(CAPTION_SPACE_URL)
|
83 |
|
84 |
-
# Assuming the caption space is public
|
85 |
caption_client = Client(CAPTION_SPACE_URL)
|
86 |
print("Gradio client initialized successfully.")
|
87 |
except Exception as e:
|
88 |
print(f"Error initializing Gradio client for {CAPTION_SPACE_URL}: {e}")
|
89 |
-
# Set client to None so the endpoint can check and return an error
|
90 |
caption_client = None
|
91 |
|
92 |
|
93 |
# Load models and initialize clients when the app starts
|
94 |
@app.on_event("startup")
|
95 |
async def startup_event():
|
96 |
-
# Load the language model (TinyLlama or Gemma)
|
97 |
load_language_model()
|
98 |
-
# Initialize the client for the captioning Space
|
99 |
initialize_caption_client()
|
100 |
|
101 |
|
102 |
-
# --- Image Captioning Function (Using gradio_client) ---
|
103 |
def generate_image_caption(image_file: UploadFile):
|
104 |
"""
|
105 |
Generates a caption for the uploaded image using the external Gradio Space API.
|
106 |
-
|
107 |
"""
|
108 |
if caption_client is None:
|
109 |
error_msg = "Gradio caption client not initialized. Cannot generate caption."
|
110 |
-
print(error_msg)
|
111 |
-
return f"Error: {error_msg}"
|
|
|
|
|
112 |
|
113 |
try:
|
114 |
print(f"Calling caption API /predict for file {image_file.filename}...")
|
115 |
|
116 |
-
# Read the content of the uploaded file
|
117 |
-
|
118 |
-
image_file.file.seek(0)
|
119 |
image_bytes = image_file.file.read()
|
120 |
|
121 |
-
#
|
122 |
-
#
|
123 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
124 |
|
125 |
# Call the predict method on the initialized client with the prepared input
|
126 |
caption = caption_client.predict(img=prepared_input, api_name="/predict")
|
127 |
|
128 |
print(f"Caption generated: {caption}")
|
129 |
return caption # Return the successful caption string
|
|
|
130 |
except Exception as e:
|
131 |
-
# Catch potential exceptions
|
132 |
-
print(f"Error during caption generation API call: {e}")
|
133 |
-
# Return an informative error string including the exception type and message
|
134 |
return f"Error: Unable to generate caption from API. Details: {type(e).__name__}: {e}"
|
135 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
136 |
|
137 |
-
# --- Language Model Story Generation Function ---
|
138 |
-
# This function uses the loaded TinyLlama model to generate the story.
|
139 |
def generate_story_tinyllama(prompt_text: str, max_new_tokens: int = 300, temperature: float = 0.7, top_p: float = 0.9, top_k: int = 50) -> str:
|
140 |
"""
|
141 |
Generates text using the loaded TinyLlama model based on the prompt.
|
@@ -144,117 +156,56 @@ def generate_story_tinyllama(prompt_text: str, max_new_tokens: int = 300, temper
|
|
144 |
if tokenizer is None or model is None:
|
145 |
raise RuntimeError("Language model and tokenizer not loaded. Cannot generate story.")
|
146 |
|
147 |
-
# TinyLlama-Chat uses a chat template (similar to Llama/Gemma's instruction format)
|
148 |
messages = [
|
149 |
{"role": "user", "content": prompt_text}
|
150 |
-
# Add {"role": "system", "content": "Your system prompt here"} if needed
|
151 |
]
|
152 |
|
153 |
-
# Apply the chat template to format the prompt correctly for the model
|
154 |
try:
|
155 |
input_text = tokenizer.apply_chat_template(
|
156 |
messages,
|
157 |
-
tokenize=False,
|
158 |
-
add_generation_prompt=True
|
159 |
)
|
160 |
except AttributeError:
|
161 |
-
|
162 |
-
|
163 |
-
input_text = f"<s>[INST] {prompt_text} [/INST]" # Basic Llama/TinyLlama instruction format
|
164 |
|
165 |
-
|
166 |
-
# max_length should be within the model's context window (e.g., 4096 for TinyLlama)
|
167 |
-
inputs = tokenizer(input_text, return_tensors="pt", padding=True, truncation=True, max_length=1024) # Truncate if prompt is too long
|
168 |
|
169 |
-
# Ensure input tensors are on the same device as the model (CPU by default)
|
170 |
-
# inputs = {k: v.to(model.device) for k, v in inputs.items()} # Redundant if model is on CPU
|
171 |
-
|
172 |
-
# Generate new tokens based on the input prompt
|
173 |
generate_ids = model.generate(
|
174 |
inputs.input_ids,
|
175 |
-
max_new_tokens=max_new_tokens,
|
176 |
-
do_sample=True,
|
177 |
-
temperature=temperature,
|
178 |
-
top_p=top_p,
|
179 |
-
top_k=top_k,
|
180 |
-
pad_token_id=tokenizer.pad_token_id,
|
181 |
-
# eos_token_id=tokenizer.eos_token_id # Optional: specify end-of-sequence token id to stop generation early
|
182 |
)
|
183 |
|
184 |
-
# Decode the generated token IDs back into text
|
185 |
-
# Slice [0, inputs.input_ids.shape[-1]:] to get only the newly generated tokens
|
186 |
-
# skip_special_tokens=True removes tokens like <s>, </s>, <pad>
|
187 |
generated_text = tokenizer.decode(generate_ids[0, inputs.input_ids.shape[-1]:], skip_special_tokens=True)
|
188 |
-
|
189 |
-
return generated_text.strip() # Return the generated text, stripped of leading/trailing whitespace
|
190 |
-
|
191 |
-
# --- Optional: Gemma 2B Story Generation Function (if you prefer Gemma and get access) ---
|
192 |
-
# def generate_story_gemma(prompt_text: str, max_new_tokens: int = 300, temperature: float = 0.7, top_p: float = 0.9, top_k: int = 50) -> str:
|
193 |
-
# """
|
194 |
-
# Generates text using the loaded Gemma model based on the prompt.
|
195 |
-
# Applies the Gemma-IT Chat template.
|
196 |
-
# """
|
197 |
-
# if tokenizer is None or model is None:
|
198 |
-
# raise RuntimeError("Language model and tokenizer not loaded. Cannot generate story.")
|
199 |
-
|
200 |
-
# # Gemma-IT uses a specific chat template
|
201 |
-
# messages = [
|
202 |
-
# {"role": "user", "content": prompt_text}
|
203 |
-
# # {"role": "system", "content": "Your system prompt here"} # Gemma also supports system prompts
|
204 |
-
# ]
|
205 |
-
# input_text = tokenizer.apply_chat_template(
|
206 |
-
# messages,
|
207 |
-
# tokenize=False,
|
208 |
-
# add_generation_prompt=True # Adds the assistant turn prompt token(s)
|
209 |
-
# )
|
210 |
-
|
211 |
-
# inputs = tokenizer(input_text, return_tensors="pt", padding=True, truncation=True, max_length=1024)
|
212 |
-
|
213 |
-
# generate_ids = model.generate(
|
214 |
-
# inputs.input_ids,
|
215 |
-
# max_new_tokens=max_new_tokens,
|
216 |
-
# do_sample=True,
|
217 |
-
# temperature=temperature,
|
218 |
-
# top_p=top_p,
|
219 |
-
# top_k=top_k,
|
220 |
-
# pad_token_id=tokenizer.pad_token_id,
|
221 |
-
# )
|
222 |
-
|
223 |
-
# generated_text = tokenizer.decode(generate_ids[0, inputs.input_ids.shape[-1]:], skip_special_tokens=True)
|
224 |
-
# return generated_text.strip()
|
225 |
-
|
226 |
|
227 |
# --- FastAPI Endpoint for Story Generation ---
|
228 |
@app.post("/generate-story/")
|
229 |
async def generate_story_endpoint(image_file: UploadFile = File(...), language: str = Form(...)):
|
230 |
-
#
|
231 |
-
|
232 |
-
# Choose a random theme for the story prompt
|
233 |
story_theme = random.choice([
|
234 |
-
'an adventurous journey',
|
235 |
-
'a
|
236 |
-
'a
|
237 |
-
'a magical adventure',
|
238 |
-
'a thrilling escape',
|
239 |
-
'an unexpected discovery',
|
240 |
-
'a dangerous mission',
|
241 |
-
'a romantic escapade',
|
242 |
-
'an epic battle',
|
243 |
'a journey into the unknown'
|
244 |
])
|
245 |
|
246 |
# Step 1: Get image caption using the external API via gradio_client
|
247 |
-
# Pass the UploadFile object
|
248 |
caption = generate_image_caption(image_file)
|
249 |
|
250 |
# Check if caption generation failed
|
251 |
-
if caption.startswith("Error:"):
|
252 |
-
print(f"Caption generation failed: {caption}")
|
253 |
-
# Return a 500 Internal Server Error with the error message
|
254 |
raise HTTPException(status_code=500, detail=caption)
|
255 |
|
256 |
# Step 2: Construct the prompt for the language model
|
257 |
-
# We instruct the model to write a story based on the theme and incorporate the caption.
|
258 |
prompt_text = f"Write an attractive story of around 300 words about {story_theme}. Incorporate the following details from an image description into the story: {caption}\n\nStory:"
|
259 |
|
260 |
# Step 3: Generate the story using the local language model
|
@@ -262,72 +213,50 @@ async def generate_story_endpoint(image_file: UploadFile = File(...), language:
|
|
262 |
# Call the appropriate story generation function (TinyLlama in this case)
|
263 |
story = generate_story_tinyllama(
|
264 |
prompt_text,
|
265 |
-
max_new_tokens=300,
|
266 |
-
temperature=0.7,
|
267 |
top_p=0.9,
|
268 |
top_k=50
|
269 |
)
|
270 |
-
story = story.strip()
|
271 |
|
272 |
except RuntimeError as e:
|
273 |
-
# Catch errors specifically from model not being loaded
|
274 |
print(f"Language model not loaded error: {e}")
|
275 |
raise HTTPException(status_code=503, detail=f"Language model not available: {e}")
|
276 |
except Exception as e:
|
277 |
-
|
278 |
-
print(f"Story generation failed: {e}") # Log the exception server-side
|
279 |
raise HTTPException(status_code=500, detail=f"Story generation failed: {type(e).__name__}: {e}. Please check Space logs for details.")
|
280 |
|
281 |
|
282 |
# Step 4: Translate the story if the target language is not English
|
283 |
if language and language.lower() != "english":
|
284 |
try:
|
285 |
-
# Use GoogleTranslator with specified source and target languages
|
286 |
translator = GoogleTranslator(source='english', target=language.lower())
|
287 |
translated_story = translator.translate(story)
|
288 |
|
289 |
-
# Check if translation was successful or returned None
|
290 |
if translated_story is None or translated_story == "":
|
291 |
print(f"Translation returned None or empty string for language: {language}")
|
292 |
-
# If translation fails, return the English story with a warning message
|
293 |
return {"story": story + "\n\n(Note: Automatic translation to your requested language failed.)"}
|
294 |
|
295 |
-
story = translated_story
|
296 |
|
297 |
except InvalidSourceOrTargetLanguage:
|
298 |
print(f"Invalid target language requested: {language}")
|
299 |
raise HTTPException(status_code=400, detail=f"Invalid target language: {language}")
|
300 |
except Exception as e:
|
301 |
-
|
302 |
-
print(f"Translation failed for language {language}: {e}") # Log server-side
|
303 |
raise HTTPException(status_code=500, detail=f"Translation failed: {type(e).__name__}: {e}")
|
304 |
|
305 |
-
|
306 |
-
# Step 5: Return the final generated (and potentially translated) story
|
307 |
return {"story": story}
|
308 |
|
309 |
# --- Optional: Serve a simple HTML form for testing ---
|
310 |
-
#
|
311 |
-
# Also uncomment the imports at the top related to HTMLResponse, StaticFiles, Jinja2Templates, Request.
|
312 |
# from fastapi import Request
|
313 |
# from fastapi.templating import Jinja2Templates
|
314 |
# from fastapi.staticfiles import StaticFiles
|
315 |
# templates = Jinja2Templates(directory="templates")
|
316 |
# app.mount("/static", StaticFiles(directory="static"), name="static")
|
317 |
-
|
318 |
# @app.get("/", response_class=HTMLResponse)
|
319 |
# async def read_root(request: Request):
|
320 |
-
# # Example index.html structure for a simple form:
|
321 |
-
# # <!DOCTYPE html>
|
322 |
-
# # <html>
|
323 |
-
# # <head><title>Story Generator</title></head>
|
324 |
-
# # <body>
|
325 |
-
# # <h1>Generate a Story from an Image</h1>
|
326 |
-
# # <form action="/generate-story/" method="post" enctype="multipart/form-data">
|
327 |
-
# # <input type="file" name="image_file" accept="image/*" required><br><br>
|
328 |
-
# # Target Language (e.g., english, french, spanish): <input type="text" name="language" value="english"><br><br>
|
329 |
-
# # <button type="submit">Generate Story</button>
|
330 |
-
# # </form>
|
331 |
-
# # </body>
|
332 |
-
# # </html>
|
333 |
# return templates.TemplateResponse("index.html", {"request": request})
|
|
|
5 |
# from fastapi.templating import Jinja2Templates
|
6 |
# from fastapi.responses import FileResponse
|
7 |
|
8 |
+
# Removed 'requests'
|
9 |
# import requests
|
10 |
+
import base64 # Keep if needed elsewhere
|
11 |
import os
|
12 |
import random
|
13 |
+
from typing import IO
|
14 |
|
15 |
# Import necessary classes from transformers
|
16 |
import torch
|
|
|
19 |
# Import the Gradio Client and handle_file
|
20 |
from gradio_client import Client, handle_file
|
21 |
|
22 |
+
# Import necessary modules for temporary file handling
|
23 |
+
import tempfile
|
24 |
+
import shutil # Useful for potential cleanup, although os.remove is sufficient here
|
25 |
+
|
26 |
+
|
27 |
from deep_translator import GoogleTranslator
|
28 |
from deep_translator.exceptions import InvalidSourceOrTargetLanguage
|
29 |
|
|
|
31 |
app = FastAPI()
|
32 |
|
33 |
# --- Hugging Face Language Model Setup (Local Inference) ---
|
34 |
+
# Using TinyLlama 1.1B Chat as the example
|
|
|
|
|
|
|
35 |
model_name = "TinyLlama/TinyLlama-1.1B-Chat-v1.0"
|
36 |
tokenizer = None
|
37 |
model = None
|
|
|
49 |
print(f"Loading language model: {model_name}...")
|
50 |
try:
|
51 |
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
|
|
52 |
if tokenizer.pad_token is None:
|
53 |
tokenizer.pad_token = tokenizer.eos_token
|
54 |
|
|
|
55 |
model = AutoModelForCausalLM.from_pretrained(
|
56 |
model_name,
|
57 |
torch_dtype=torch.float16, # Use float16 precision
|
|
|
58 |
)
|
|
|
59 |
|
60 |
print(f"Language model {model_name} loaded successfully.")
|
61 |
except Exception as e:
|
62 |
print(f"Error loading language model {model_name}: {e}")
|
|
|
|
|
63 |
tokenizer = None
|
64 |
model = None
|
65 |
|
|
|
69 |
global caption_client
|
70 |
print(f"Initializing Gradio client for {CAPTION_SPACE_URL}...")
|
71 |
try:
|
72 |
+
# If the target Space is private or requires authentication,
|
73 |
+
# uncomment and set HF_TOKEN as a Space Secret.
|
74 |
# HF_TOKEN = os.environ.get("HF_TOKEN")
|
75 |
# if HF_TOKEN:
|
76 |
# caption_client = Client(CAPTION_SPACE_URL, hf_token=HF_TOKEN)
|
77 |
# else:
|
78 |
# caption_client = Client(CAPTION_SPACE_URL)
|
79 |
|
80 |
+
# Assuming the caption space is public
|
81 |
caption_client = Client(CAPTION_SPACE_URL)
|
82 |
print("Gradio client initialized successfully.")
|
83 |
except Exception as e:
|
84 |
print(f"Error initializing Gradio client for {CAPTION_SPACE_URL}: {e}")
|
|
|
85 |
caption_client = None
|
86 |
|
87 |
|
88 |
# Load models and initialize clients when the app starts
|
89 |
@app.on_event("startup")
|
90 |
async def startup_event():
|
|
|
91 |
load_language_model()
|
|
|
92 |
initialize_caption_client()
|
93 |
|
94 |
|
95 |
+
# --- Image Captioning Function (Using gradio_client and temporary file) ---
|
96 |
def generate_image_caption(image_file: UploadFile):
|
97 |
"""
|
98 |
Generates a caption for the uploaded image using the external Gradio Space API.
|
99 |
+
Saves the uploaded file to a temporary file and uses its path with handle_file.
|
100 |
"""
|
101 |
if caption_client is None:
|
102 |
error_msg = "Gradio caption client not initialized. Cannot generate caption."
|
103 |
+
print(error_msg)
|
104 |
+
return f"Error: {error_msg}"
|
105 |
+
|
106 |
+
temp_file_path = None # Initialize temporary file path variable
|
107 |
|
108 |
try:
|
109 |
print(f"Calling caption API /predict for file {image_file.filename}...")
|
110 |
|
111 |
+
# Read the content of the uploaded file
|
112 |
+
image_file.file.seek(0) # Ensure pointer is at the beginning
|
|
|
113 |
image_bytes = image_file.file.read()
|
114 |
|
115 |
+
# Create a temporary file and write the bytes to it
|
116 |
+
# use delete=False so the file isn't deleted automatically when closed
|
117 |
+
# Use suffix to hint at the file type
|
118 |
+
temp_file = tempfile.NamedTemporaryFile(delete=False, suffix=os.path.splitext(image_file.filename)[1] or '.jpg')
|
119 |
+
temp_file.write(image_bytes)
|
120 |
+
temp_file.close() # Close the handle so gradio_client can access it
|
121 |
+
temp_file_path = temp_file.name # Get the actual path string
|
122 |
+
|
123 |
+
print(f"Saved uploaded file temporarily to: {temp_file_path}")
|
124 |
+
|
125 |
+
# Use handle_file() with the path to the temporary file (a string)
|
126 |
+
# This aligns with the documentation "img filepath Required" and example handle_file('...')
|
127 |
+
prepared_input = handle_file(temp_file_path)
|
128 |
|
129 |
# Call the predict method on the initialized client with the prepared input
|
130 |
caption = caption_client.predict(img=prepared_input, api_name="/predict")
|
131 |
|
132 |
print(f"Caption generated: {caption}")
|
133 |
return caption # Return the successful caption string
|
134 |
+
|
135 |
except Exception as e:
|
136 |
+
# Catch potential exceptions during the process
|
137 |
+
print(f"Error during caption generation API call: {e}")
|
|
|
138 |
return f"Error: Unable to generate caption from API. Details: {type(e).__name__}: {e}"
|
139 |
|
140 |
+
finally:
|
141 |
+
# Clean up the temporary file
|
142 |
+
if temp_file_path and os.path.exists(temp_file_path):
|
143 |
+
print(f"Cleaning up temporary file: {temp_file_path}")
|
144 |
+
try:
|
145 |
+
os.remove(temp_file_path)
|
146 |
+
except OSError as e:
|
147 |
+
print(f"Error removing temporary file {temp_file_path}: {e}") # Log cleanup errors
|
148 |
+
|
149 |
|
150 |
+
# --- Language Model Story Generation Function (TinyLlama) ---
|
|
|
151 |
def generate_story_tinyllama(prompt_text: str, max_new_tokens: int = 300, temperature: float = 0.7, top_p: float = 0.9, top_k: int = 50) -> str:
|
152 |
"""
|
153 |
Generates text using the loaded TinyLlama model based on the prompt.
|
|
|
156 |
if tokenizer is None or model is None:
|
157 |
raise RuntimeError("Language model and tokenizer not loaded. Cannot generate story.")
|
158 |
|
|
|
159 |
messages = [
|
160 |
{"role": "user", "content": prompt_text}
|
|
|
161 |
]
|
162 |
|
|
|
163 |
try:
|
164 |
input_text = tokenizer.apply_chat_template(
|
165 |
messages,
|
166 |
+
tokenize=False,
|
167 |
+
add_generation_prompt=True
|
168 |
)
|
169 |
except AttributeError:
|
170 |
+
print("Warning: apply_chat_template not found. Using basic prompt formatting.")
|
171 |
+
input_text = f"<s>[INST] {prompt_text} [/INST]"
|
|
|
172 |
|
173 |
+
inputs = tokenizer(input_text, return_tensors="pt", padding=True, truncation=True, max_length=1024)
|
|
|
|
|
174 |
|
|
|
|
|
|
|
|
|
175 |
generate_ids = model.generate(
|
176 |
inputs.input_ids,
|
177 |
+
max_new_tokens=max_new_tokens,
|
178 |
+
do_sample=True,
|
179 |
+
temperature=temperature,
|
180 |
+
top_p=top_p,
|
181 |
+
top_k=top_k,
|
182 |
+
pad_token_id=tokenizer.pad_token_id,
|
|
|
183 |
)
|
184 |
|
|
|
|
|
|
|
185 |
generated_text = tokenizer.decode(generate_ids[0, inputs.input_ids.shape[-1]:], skip_special_tokens=True)
|
186 |
+
return generated_text.strip()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
187 |
|
188 |
# --- FastAPI Endpoint for Story Generation ---
|
189 |
@app.post("/generate-story/")
|
190 |
async def generate_story_endpoint(image_file: UploadFile = File(...), language: str = Form(...)):
|
191 |
+
# Choose a random theme
|
|
|
|
|
192 |
story_theme = random.choice([
|
193 |
+
'an adventurous journey', 'a mysterious encounter', 'a heroic quest',
|
194 |
+
'a magical adventure', 'a thrilling escape', 'an unexpected discovery',
|
195 |
+
'a dangerous mission', 'a romantic escapade', 'an epic battle',
|
|
|
|
|
|
|
|
|
|
|
|
|
196 |
'a journey into the unknown'
|
197 |
])
|
198 |
|
199 |
# Step 1: Get image caption using the external API via gradio_client
|
200 |
+
# Pass the UploadFile object to the captioning function
|
201 |
caption = generate_image_caption(image_file)
|
202 |
|
203 |
# Check if caption generation failed
|
204 |
+
if caption.startswith("Error:"):
|
205 |
+
print(f"Caption generation failed: {caption}")
|
|
|
206 |
raise HTTPException(status_code=500, detail=caption)
|
207 |
|
208 |
# Step 2: Construct the prompt for the language model
|
|
|
209 |
prompt_text = f"Write an attractive story of around 300 words about {story_theme}. Incorporate the following details from an image description into the story: {caption}\n\nStory:"
|
210 |
|
211 |
# Step 3: Generate the story using the local language model
|
|
|
213 |
# Call the appropriate story generation function (TinyLlama in this case)
|
214 |
story = generate_story_tinyllama(
|
215 |
prompt_text,
|
216 |
+
max_new_tokens=300,
|
217 |
+
temperature=0.7,
|
218 |
top_p=0.9,
|
219 |
top_k=50
|
220 |
)
|
221 |
+
story = story.strip()
|
222 |
|
223 |
except RuntimeError as e:
|
|
|
224 |
print(f"Language model not loaded error: {e}")
|
225 |
raise HTTPException(status_code=503, detail=f"Language model not available: {e}")
|
226 |
except Exception as e:
|
227 |
+
print(f"Story generation failed: {e}")
|
|
|
228 |
raise HTTPException(status_code=500, detail=f"Story generation failed: {type(e).__name__}: {e}. Please check Space logs for details.")
|
229 |
|
230 |
|
231 |
# Step 4: Translate the story if the target language is not English
|
232 |
if language and language.lower() != "english":
|
233 |
try:
|
|
|
234 |
translator = GoogleTranslator(source='english', target=language.lower())
|
235 |
translated_story = translator.translate(story)
|
236 |
|
|
|
237 |
if translated_story is None or translated_story == "":
|
238 |
print(f"Translation returned None or empty string for language: {language}")
|
|
|
239 |
return {"story": story + "\n\n(Note: Automatic translation to your requested language failed.)"}
|
240 |
|
241 |
+
story = translated_story
|
242 |
|
243 |
except InvalidSourceOrTargetLanguage:
|
244 |
print(f"Invalid target language requested: {language}")
|
245 |
raise HTTPException(status_code=400, detail=f"Invalid target language: {language}")
|
246 |
except Exception as e:
|
247 |
+
print(f"Translation failed for language {language}: {e}")
|
|
|
248 |
raise HTTPException(status_code=500, detail=f"Translation failed: {type(e).__name__}: {e}")
|
249 |
|
250 |
+
# Step 5: Return the final story
|
|
|
251 |
return {"story": story}
|
252 |
|
253 |
# --- Optional: Serve a simple HTML form for testing ---
|
254 |
+
# Requires uncommenting imports above and creating 'templates/index.html'
|
|
|
255 |
# from fastapi import Request
|
256 |
# from fastapi.templating import Jinja2Templates
|
257 |
# from fastapi.staticfiles import StaticFiles
|
258 |
# templates = Jinja2Templates(directory="templates")
|
259 |
# app.mount("/static", StaticFiles(directory="static"), name="static")
|
|
|
260 |
# @app.get("/", response_class=HTMLResponse)
|
261 |
# async def read_root(request: Request):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
262 |
# return templates.TemplateResponse("index.html", {"request": request})
|