Anaconda024 commited on
Commit
84cd483
·
verified ·
1 Parent(s): 3a3bc2d

Qwen check if model is accessible

Browse files
Files changed (1) hide show
  1. app.py +87 -15
app.py CHANGED
@@ -126,6 +126,22 @@ def build_index():
126
  index = faiss.IndexFlatL2(384) # Dummy index with correct dim for all-MiniLM-L6-v2
127
 
128
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
129
  def is_token_valid(token: str) -> bool:
130
  """
131
  Performs a basic check if the HF token is valid by calling the /api/whoami-v2 endpoint.
@@ -154,7 +170,7 @@ def is_token_valid(token: str) -> bool:
154
  def ask_question(question, hf_token, model_id):
155
  global index, chunks
156
  print("--- Starting ask_question function ---")
157
-
158
  # 1. Check index
159
  print(f"1. Checking index and chunks...")
160
  if index is None:
@@ -176,9 +192,18 @@ def ask_question(question, hf_token, model_id):
176
  return error_msg
177
  print(" -> Token validated successfully.")
178
 
179
- # 3. Process Question
180
- print(f"3. Processing question: '{question}' using model '{model_id}'...")
181
-
 
 
 
 
 
 
 
 
 
182
  # --- Detailed Steps with Logging ---
183
  print(" a. Encoding question...")
184
  q_emb = embedder.encode([question], convert_to_numpy=True)
@@ -203,22 +228,70 @@ def ask_question(question, hf_token, model_id):
203
  print(" f. Constructing prompt...")
204
  prompt = f"Answer based on context:\n{context}\nQuestion: {question}\nAnswer:"
205
  print(f" -> Prompt constructed. Length: {len(prompt)} characters.")
206
- # Optional: Print first part of prompt for debugging
207
- # print(f" -> Prompt preview: {prompt[:200]}...")
208
 
209
- # 4. Call Model API
210
- print("4. Initializing InferenceClient...")
 
211
  client = InferenceClient(model=model_id, token=hf_token)
212
  print(" -> InferenceClient initialized.")
213
 
214
- print("5. Calling model API (text_generation)...")
215
- # Using text_generation as it's the standard method
216
- resp = client.text_generation(prompt, max_new_tokens=200, temperature=0.7)
217
- print(" -> Model API call successful.")
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
218
 
219
  print("--- ask_question function completed successfully ---")
220
  return resp
221
 
 
 
 
 
222
  except Exception as e:
223
  error_msg = f"Error processing question: {str(e)}"
224
  print(f" -> {error_msg}")
@@ -227,7 +300,6 @@ def ask_question(question, hf_token, model_id):
227
  traceback.print_exc()
228
  print("--- ask_question function failed ---")
229
  return error_msg # Return the error message to the UI
230
-
231
  # ... (rest of the code remains the same)
232
  # ==========================
233
  # Build index on startup
@@ -262,7 +334,7 @@ with gr.Blocks(title="UCC RAG Chatbot") as demo:
262
  with gr.Column(scale=1):
263
  gr.Markdown("### Configuration")
264
  hf_token = gr.Textbox(label="Hugging Face Token", type="password", info="Get your token from https://huggingface.co/settings/tokens")
265
- model_id = gr.Textbox(label="Model ID", value="google/flan-t5-base", info="ID of the Hugging Face model to use")
266
  with gr.Column(scale=2):
267
  gr.Markdown("### Ask a Question")
268
  question = gr.Textbox(label="Your Question", placeholder="e.g., What is the role of UCC?")
@@ -280,7 +352,7 @@ with gr.Blocks(title="UCC RAG Chatbot") as demo:
280
  # inputs=[
281
  # gr.Textbox(label="Question"),
282
  # gr.Textbox(label="HF Token", type="password"),
283
- # gr.Textbox(label="Model ID", value="google/flan-t5-base")
284
  # ],
285
  # outputs=gr.JSON(),
286
  # title="API Endpoint",
 
126
  index = faiss.IndexFlatL2(384) # Dummy index with correct dim for all-MiniLM-L6-v2
127
 
128
 
129
+ def is_model_accessible(token: str, model_id: str) -> bool:
130
+ """
131
+ Checks if the model exists and if the user has permission to access it via the Hub API.
132
+ """
133
+ try:
134
+ api = HfApi()
135
+ # This call requires a valid token and checks permissions
136
+ model_info = api.model_info(model_id, token=token)
137
+ print(f"Model '{model_id}' is accessible. Model info retrieved.")
138
+ # print(f"Model info snippet: {model_info}") # Uncomment for more details
139
+ return True
140
+ except Exception as e:
141
+ print(f"Model accessibility check failed for '{model_id}': {e}")
142
+ # Check specific error types if needed (e.g., HTTP 401, 403, 404)
143
+ return False
144
+
145
  def is_token_valid(token: str) -> bool:
146
  """
147
  Performs a basic check if the HF token is valid by calling the /api/whoami-v2 endpoint.
 
170
  def ask_question(question, hf_token, model_id):
171
  global index, chunks
172
  print("--- Starting ask_question function ---")
173
+
174
  # 1. Check index
175
  print(f"1. Checking index and chunks...")
176
  if index is None:
 
192
  return error_msg
193
  print(" -> Token validated successfully.")
194
 
195
+ # 3. Check Model Accessibility
196
+ print(f"3. Checking model accessibility: '{model_id}'...")
197
+ if not is_model_accessible(hf_token, model_id):
198
+ error_msg = f"Model '{model_id}' is not accessible with the provided token or does not exist. Please check the model ID and token permissions."
199
+ print(f" -> {error_msg}")
200
+ return error_msg
201
+ print(" -> Model accessibility check passed.")
202
+
203
+
204
+ # 4. Process Question
205
+ print(f"4. Processing question: '{question}' using model '{model_id}'...")
206
+
207
  # --- Detailed Steps with Logging ---
208
  print(" a. Encoding question...")
209
  q_emb = embedder.encode([question], convert_to_numpy=True)
 
228
  print(" f. Constructing prompt...")
229
  prompt = f"Answer based on context:\n{context}\nQuestion: {question}\nAnswer:"
230
  print(f" -> Prompt constructed. Length: {len(prompt)} characters.")
 
 
231
 
232
+
233
+ # 5. Call Model API
234
+ print("5. Initializing InferenceClient...")
235
  client = InferenceClient(model=model_id, token=hf_token)
236
  print(" -> InferenceClient initialized.")
237
 
238
+ print("6. Calling model API (text_generation)...")
239
+
240
+ # --- Potential Workaround ---
241
+ # Instead of letting the client auto-detect, explicitly specify the endpoint URL
242
+ # This bypasses some of the internal routing logic that might be failing.
243
+ try:
244
+ # Try the standard call first
245
+ resp = client.text_generation(prompt, max_new_tokens=200, temperature=0.7)
246
+ print(" -> Model API call (standard) successful.")
247
+ except StopIteration:
248
+ # If the specific error occurs, try the workaround
249
+ print(" -> Standard call failed with StopIteration. Trying direct endpoint URL...")
250
+ # Construct the direct Inference API endpoint URL for the model
251
+ api_url = f"https://api-inference.huggingface.co/models/{model_id}"
252
+ headers = {"Authorization": f"Bearer {hf_token}"}
253
+ payload = {
254
+ "inputs": prompt,
255
+ "parameters": {
256
+ "max_new_tokens": 200,
257
+ "temperature": 0.7,
258
+ # Add other parameters if needed
259
+ }
260
+ }
261
+ print(f" -> Sending POST request to {api_url}")
262
+ response = requests.post(api_url, headers=headers, json=payload)
263
+ response.raise_for_status() # Raise an exception for bad status codes (4xx or 5xx)
264
+ # Handle the response based on the model type
265
+ # For text2text-generation models like Flan-T5, the response is usually JSON with a 'generated_text' key
266
+ # For text-generation models, it might be just the text or a list
267
+ # Let's try parsing as JSON first
268
+ try:
269
+ result = response.json()
270
+ if isinstance(result, list) and len(result) > 0:
271
+ # Often the first item is a dict with 'generated_text'
272
+ if isinstance(result[0], dict) and 'generated_text' in result[0]:
273
+ resp = result[0]['generated_text']
274
+ else:
275
+ # If it's a list of strings or other format, adapt here
276
+ resp = str(result[0])
277
+ elif isinstance(result, dict) and 'generated_text' in result:
278
+ resp = result['generated_text']
279
+ else:
280
+ # Fallback: convert the whole response to string
281
+ resp = str(result)
282
+ except requests.exceptions.JSONDecodeError:
283
+ # If not JSON, assume raw text (less common for this endpoint but possible)
284
+ resp = response.text
285
+
286
+ print(" -> Model API call (direct) successful.")
287
 
288
  print("--- ask_question function completed successfully ---")
289
  return resp
290
 
291
+ except requests.exceptions.RequestException as e: # Catch network-related errors from requests
292
+ error_msg = f"Network error calling model API: {e}"
293
+ print(f" -> {error_msg}")
294
+ return error_msg
295
  except Exception as e:
296
  error_msg = f"Error processing question: {str(e)}"
297
  print(f" -> {error_msg}")
 
300
  traceback.print_exc()
301
  print("--- ask_question function failed ---")
302
  return error_msg # Return the error message to the UI
 
303
  # ... (rest of the code remains the same)
304
  # ==========================
305
  # Build index on startup
 
334
  with gr.Column(scale=1):
335
  gr.Markdown("### Configuration")
336
  hf_token = gr.Textbox(label="Hugging Face Token", type="password", info="Get your token from https://huggingface.co/settings/tokens")
337
+ model_id = gr.Textbox(label="Model ID", value="deepseek-llm-7b-base", info="ID of the Hugging Face model to use")
338
  with gr.Column(scale=2):
339
  gr.Markdown("### Ask a Question")
340
  question = gr.Textbox(label="Your Question", placeholder="e.g., What is the role of UCC?")
 
352
  # inputs=[
353
  # gr.Textbox(label="Question"),
354
  # gr.Textbox(label="HF Token", type="password"),
355
+ # gr.Textbox(label="Model ID", value="deepseek-llm-7b-base")
356
  # ],
357
  # outputs=gr.JSON(),
358
  # title="API Endpoint",