Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -134,39 +134,62 @@ class Transformer(nn.Module):
|
|
134 |
|
135 |
# Set device
|
136 |
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
|
|
137 |
|
138 |
# Load tokenizers
|
139 |
-
|
140 |
-
|
141 |
-
|
142 |
-
|
143 |
-
|
144 |
-
|
145 |
-
|
146 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
147 |
|
148 |
def generate_pseudocode(cpp_code, max_len):
|
149 |
"""Generate pseudocode from C++ code with streaming output."""
|
|
|
150 |
model.eval()
|
151 |
-
src = torch.tensor([sp_code.encode_as_ids(cpp_code)], dtype=torch.long, device=device) # Tokenize C++ code
|
152 |
-
tgt = torch.tensor([[2]], dtype=torch.long, device=device) # <bos_id>=2
|
153 |
|
154 |
-
|
155 |
-
|
156 |
-
|
157 |
-
|
158 |
-
|
159 |
-
|
160 |
-
|
161 |
-
|
162 |
-
|
163 |
-
|
164 |
-
|
165 |
-
|
166 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
167 |
|
168 |
def respond(message, history, max_tokens):
|
169 |
"""Wrapper for Gradio interface."""
|
|
|
170 |
for response in generate_pseudocode(message, max_tokens):
|
171 |
yield response
|
172 |
|
@@ -183,4 +206,4 @@ demo = gr.ChatInterface(
|
|
183 |
)
|
184 |
|
185 |
if __name__ == "__main__":
|
186 |
-
demo.launch()
|
|
|
134 |
|
135 |
# Set device
|
136 |
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
137 |
+
print(f"Using device: {device}")
|
138 |
|
139 |
# Load tokenizers
|
140 |
+
try:
|
141 |
+
sp_pseudo = spm.SentencePieceProcessor(model_file="pseudo.model")
|
142 |
+
sp_code = spm.SentencePieceProcessor(model_file="code.model")
|
143 |
+
print("Tokenizers loaded successfully.")
|
144 |
+
except Exception as e:
|
145 |
+
print(f"Error loading tokenizers: {e}")
|
146 |
+
raise
|
147 |
+
|
148 |
+
# Load the full saved model
|
149 |
+
model_path = "transformer_cpp_to_pseudo_30.pth"
|
150 |
+
try:
|
151 |
+
model = torch.load(model_path, map_location=device, weights_only=False)
|
152 |
+
model.eval()
|
153 |
+
model = model.to(device)
|
154 |
+
print("Model loaded successfully.")
|
155 |
+
except Exception as e:
|
156 |
+
print(f"Error loading model: {e}")
|
157 |
+
raise
|
158 |
|
159 |
def generate_pseudocode(cpp_code, max_len):
|
160 |
"""Generate pseudocode from C++ code with streaming output."""
|
161 |
+
print(f"Input C++ code: {cpp_code}")
|
162 |
model.eval()
|
|
|
|
|
163 |
|
164 |
+
try:
|
165 |
+
src_tokens = sp_code.encode_as_ids(cpp_code)
|
166 |
+
print(f"Source tokens: {src_tokens}")
|
167 |
+
src = torch.tensor([src_tokens], dtype=torch.long, device=device)
|
168 |
+
|
169 |
+
tgt = torch.tensor([[2]], dtype=torch.long, device=device) # <bos_id>=2
|
170 |
+
generated_tokens = [2] # Start with <START>
|
171 |
+
response = ""
|
172 |
+
|
173 |
+
with torch.no_grad():
|
174 |
+
for i in range(max_len):
|
175 |
+
output = model(src, tgt)
|
176 |
+
next_token = output[:, -1, :].argmax(-1).item()
|
177 |
+
generated_tokens.append(next_token)
|
178 |
+
tgt = torch.cat([tgt, torch.tensor([[next_token]], device=device)], dim=1)
|
179 |
+
response = sp_pseudo.decode_ids(generated_tokens)
|
180 |
+
print(f"Step {i}: Next token = {next_token}, Generated so far: {response}")
|
181 |
+
yield response # Yield partial output
|
182 |
+
if next_token == 3: # <END>=3
|
183 |
+
print("EOS token detected, stopping generation.")
|
184 |
+
break
|
185 |
+
yield response # Final output
|
186 |
+
except Exception as e:
|
187 |
+
print(f"Error in generation: {e}")
|
188 |
+
yield f"Error: {e}"
|
189 |
|
190 |
def respond(message, history, max_tokens):
|
191 |
"""Wrapper for Gradio interface."""
|
192 |
+
print(f"Received message: {message}")
|
193 |
for response in generate_pseudocode(message, max_tokens):
|
194 |
yield response
|
195 |
|
|
|
206 |
)
|
207 |
|
208 |
if __name__ == "__main__":
|
209 |
+
demo.launch(debug=True) # Enable debug mode for more output
|