Spaces:
Sleeping
Sleeping
salomonsky
commited on
Update app.py
Browse files
app.py
CHANGED
@@ -1,15 +1,28 @@
|
|
1 |
-
from functools import lru_cache
|
2 |
-
import gradio as gr
|
3 |
-
import threading
|
4 |
-
import time
|
5 |
from huggingface_hub import InferenceClient
|
|
|
6 |
|
7 |
client = InferenceClient("mistralai/Mixtral-8x7B-Instruct-v0.1")
|
8 |
system_prompt = "Deber谩s proporcinar c贸digo limpio, resumido, investigando cada l铆nea en internet en foros, revisa stackoverflow.com para consultas, elimina comentarios siempre, concatena el funcionamiento de los bloques y bibliotecas, y esquematiza el funcionamiento global del c贸digo. Preve茅 posibles errores y complementa al final con una tabla explicando el funcionamiento, propon alternativas de bibliotecas para solucionar errores, siempre consulta internet para posibles resoluciones."
|
9 |
system_prompt_sent = False
|
10 |
|
11 |
-
|
12 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
13 |
global system_prompt_sent
|
14 |
temperature = float(temperature)
|
15 |
if temperature < 1e-2:
|
@@ -25,35 +38,17 @@ def generate_text(prompt, temperature=0.9, max_new_tokens=4096, top_p=0.95, repe
|
|
25 |
seed=42,
|
26 |
)
|
27 |
|
28 |
-
formatted_prompt = format_prompt(prompt,
|
29 |
-
|
30 |
stream = client.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=True)
|
31 |
output = ""
|
32 |
|
33 |
for response in stream:
|
34 |
output += response.token.text
|
35 |
-
|
36 |
-
system_prompt_sent = False
|
37 |
|
38 |
return output
|
39 |
|
40 |
-
def format_prompt(message, system_prompt_sent):
|
41 |
-
prompt = " "
|
42 |
-
|
43 |
-
if not system_prompt_sent:
|
44 |
-
prompt += f"[INST] {system_prompt} [/ "
|
45 |
-
system_prompt_sent = True
|
46 |
-
|
47 |
-
return prompt
|
48 |
-
|
49 |
-
def generate(
|
50 |
-
message, history, temperature=0.9, max_new_tokens=4096, top_p=0.95, repetition_penalty=1.0,
|
51 |
-
):
|
52 |
-
t = threading.Thread(target=generate_text, args=(message, temperature, max_new_tokens, top_p, repetition_penalty))
|
53 |
-
t.start()
|
54 |
-
t.join()
|
55 |
-
return generate_text.cache_info()[1], generate_text(message, temperature, max_new_tokens, top_p, repetition_penalty)
|
56 |
-
|
57 |
chat_interface = gr.ChatInterface(
|
58 |
fn=generate,
|
59 |
chatbot=gr.Chatbot(show_label=False, show_share_button=False, show_copy_button=False, likeable=False, layout="vertical", height=900),
|
|
|
|
|
|
|
|
|
|
|
1 |
from huggingface_hub import InferenceClient
|
2 |
+
import gradio as gr
|
3 |
|
4 |
client = InferenceClient("mistralai/Mixtral-8x7B-Instruct-v0.1")
|
5 |
system_prompt = "Deber谩s proporcinar c贸digo limpio, resumido, investigando cada l铆nea en internet en foros, revisa stackoverflow.com para consultas, elimina comentarios siempre, concatena el funcionamiento de los bloques y bibliotecas, y esquematiza el funcionamiento global del c贸digo. Preve茅 posibles errores y complementa al final con una tabla explicando el funcionamiento, propon alternativas de bibliotecas para solucionar errores, siempre consulta internet para posibles resoluciones."
|
6 |
system_prompt_sent = False
|
7 |
|
8 |
+
def format_prompt(message, history):
|
9 |
+
global system_prompt_sent
|
10 |
+
prompt = "<s>"
|
11 |
+
|
12 |
+
if not any(f"[INST] {system_prompt} [/INST]" in user_prompt for user_prompt, _ in history):
|
13 |
+
prompt += f"[INST] {system_prompt} [/INST]"
|
14 |
+
system_prompt_sent = True
|
15 |
+
|
16 |
+
for user_prompt, bot_response in history:
|
17 |
+
prompt += f"[INST] {user_prompt} [/INST]"
|
18 |
+
prompt += f" {bot_response}</s> "
|
19 |
+
|
20 |
+
prompt += f"[INST] {message} [/INST]"
|
21 |
+
return prompt
|
22 |
+
|
23 |
+
def generate(
|
24 |
+
prompt, history, temperature=0.9, max_new_tokens=4096, top_p=0.95, repetition_penalty=1.0,
|
25 |
+
):
|
26 |
global system_prompt_sent
|
27 |
temperature = float(temperature)
|
28 |
if temperature < 1e-2:
|
|
|
38 |
seed=42,
|
39 |
)
|
40 |
|
41 |
+
formatted_prompt = format_prompt(prompt, history)
|
42 |
+
|
43 |
stream = client.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=True)
|
44 |
output = ""
|
45 |
|
46 |
for response in stream:
|
47 |
output += response.token.text
|
48 |
+
yield output
|
|
|
49 |
|
50 |
return output
|
51 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
52 |
chat_interface = gr.ChatInterface(
|
53 |
fn=generate,
|
54 |
chatbot=gr.Chatbot(show_label=False, show_share_button=False, show_copy_button=False, likeable=False, layout="vertical", height=900),
|