Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -15,31 +15,31 @@ tokenizer, yi_coder_model, yi_coder_device = load_yi_coder_model()
|
|
15 |
# Conectar a Pinecone
|
16 |
index = connect_to_pinecone()
|
17 |
|
18 |
-
# Funci贸n para generar c贸digo
|
19 |
@gpu_decorator(duration=100)
|
20 |
def generate_code(system_prompt, user_prompt, max_length):
|
21 |
device = yi_coder_device
|
22 |
model = yi_coder_model
|
23 |
-
tokenizer_ = tokenizer
|
24 |
|
25 |
-
|
26 |
-
|
27 |
-
{"role": "user", "content": user_prompt}
|
28 |
-
]
|
29 |
|
30 |
-
#
|
31 |
-
prompt = system_prompt + "\n" + user_prompt
|
32 |
model_inputs = tokenizer_(prompt, return_tensors="pt").to(device)
|
33 |
|
34 |
with torch.no_grad():
|
|
|
35 |
generated_ids = model.generate(
|
36 |
model_inputs.input_ids,
|
37 |
max_new_tokens=max_length,
|
38 |
-
eos_token_id=tokenizer_.eos_token_id
|
39 |
)
|
40 |
|
41 |
-
#
|
42 |
generated_text = tokenizer_.batch_decode(generated_ids, skip_special_tokens=True)[0]
|
|
|
|
|
43 |
response = generated_text[len(prompt):].strip()
|
44 |
|
45 |
return response
|
|
|
15 |
# Conectar a Pinecone
|
16 |
index = connect_to_pinecone()
|
17 |
|
18 |
+
# Funci贸n para generar c贸digo con Yi-Coder
|
19 |
@gpu_decorator(duration=100)
|
20 |
def generate_code(system_prompt, user_prompt, max_length):
|
21 |
device = yi_coder_device
|
22 |
model = yi_coder_model
|
23 |
+
tokenizer_ = tokenizer
|
24 |
|
25 |
+
# Combina el system_prompt y user_prompt sin formato de chat
|
26 |
+
prompt = f"{system_prompt}\n{user_prompt}"
|
|
|
|
|
27 |
|
28 |
+
# Tokeniza el prompt
|
|
|
29 |
model_inputs = tokenizer_(prompt, return_tensors="pt").to(device)
|
30 |
|
31 |
with torch.no_grad():
|
32 |
+
# Genera la respuesta
|
33 |
generated_ids = model.generate(
|
34 |
model_inputs.input_ids,
|
35 |
max_new_tokens=max_length,
|
36 |
+
eos_token_id=tokenizer_.eos_token_id
|
37 |
)
|
38 |
|
39 |
+
# Decodifica el texto generado
|
40 |
generated_text = tokenizer_.batch_decode(generated_ids, skip_special_tokens=True)[0]
|
41 |
+
|
42 |
+
# Extrae solo la parte generada despu茅s del prompt inicial
|
43 |
response = generated_text[len(prompt):].strip()
|
44 |
|
45 |
return response
|