Spaces:
Sleeping
Sleeping
jordigonzm
commited on
Commit
•
31cc7d7
1
Parent(s):
1f24ef3
app.py cambios gradio
Browse files
app.py
CHANGED
@@ -1,11 +1,15 @@
|
|
1 |
import os
|
2 |
from threading import Thread
|
3 |
from typing import Iterator
|
4 |
-
|
5 |
import gradio as gr
|
6 |
import spaces
|
7 |
import torch
|
8 |
from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
|
|
|
|
|
|
|
|
|
|
|
9 |
|
10 |
MAX_MAX_NEW_TOKENS = 2048
|
11 |
DEFAULT_MAX_NEW_TOKENS = 1024
|
@@ -17,10 +21,10 @@ DESCRIPTION = """\
|
|
17 |
|
18 |
This Space demonstrates model [Llama-2-13b-chat](https://huggingface.co/meta-llama/Llama-2-13b-chat) by Meta, a Llama 2 model with 13B parameters fine-tuned for chat instructions. Feel free to play with it, or duplicate to run generations without a queue! If you want to run your own service, you can also [deploy the model on Inference Endpoints](https://huggingface.co/inference-endpoints).
|
19 |
|
20 |
-
|
21 |
|
22 |
-
|
23 |
-
|
24 |
|
25 |
"""
|
26 |
|
@@ -28,21 +32,25 @@ LICENSE = """
|
|
28 |
<p/>
|
29 |
|
30 |
---
|
31 |
-
|
32 |
-
|
33 |
"""
|
34 |
|
35 |
if not torch.cuda.is_available():
|
36 |
-
DESCRIPTION += "\n<p>Running on CPU
|
37 |
|
38 |
|
39 |
-
|
|
|
|
|
|
|
|
|
|
|
40 |
model_id = "meta-llama/Llama-2-13b-chat-hf"
|
41 |
model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto", load_in_4bit=True)
|
42 |
tokenizer = AutoTokenizer.from_pretrained(model_id)
|
43 |
tokenizer.use_default_system_prompt = False
|
44 |
|
45 |
-
|
46 |
@spaces.GPU
|
47 |
def generate(
|
48 |
message: str,
|
@@ -88,6 +96,9 @@ def generate(
|
|
88 |
yield "".join(outputs)
|
89 |
|
90 |
|
|
|
|
|
|
|
91 |
chat_interface = gr.ChatInterface(
|
92 |
fn=generate,
|
93 |
additional_inputs=[
|
@@ -128,14 +139,6 @@ chat_interface = gr.ChatInterface(
|
|
128 |
value=1.2,
|
129 |
),
|
130 |
],
|
131 |
-
stop_btn=None,
|
132 |
-
examples=[
|
133 |
-
["Hello there! How are you doing?"],
|
134 |
-
["Can you explain briefly to me what is the Python programming language?"],
|
135 |
-
["Explain the plot of Cinderella in a sentence."],
|
136 |
-
["How many hours does it take a man to eat a Helicopter?"],
|
137 |
-
["Write a 100-word article on 'Benefits of Open-Source in AI research'"],
|
138 |
-
],
|
139 |
cache_examples=False,
|
140 |
)
|
141 |
|
|
|
1 |
import os
|
2 |
from threading import Thread
|
3 |
from typing import Iterator
|
|
|
4 |
import gradio as gr
|
5 |
import spaces
|
6 |
import torch
|
7 |
from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
|
8 |
+
import huggingface_hub
|
9 |
+
|
10 |
+
# Obtener token de Hugging Face
|
11 |
+
token = os.environ.get("HUGGINGFACE_HUB_TOKEN", None)
|
12 |
+
huggingface_hub.login(token=token)
|
13 |
|
14 |
MAX_MAX_NEW_TOKENS = 2048
|
15 |
DEFAULT_MAX_NEW_TOKENS = 1024
|
|
|
21 |
|
22 |
This Space demonstrates model [Llama-2-13b-chat](https://huggingface.co/meta-llama/Llama-2-13b-chat) by Meta, a Llama 2 model with 13B parameters fine-tuned for chat instructions. Feel free to play with it, or duplicate to run generations without a queue! If you want to run your own service, you can also [deploy the model on Inference Endpoints](https://huggingface.co/inference-endpoints).
|
23 |
|
24 |
+
Para más detalles sobre la familia de modelos Llama 2 y cómo usarlos con `transformers`, echa un vistazo [a nuestro post de blog](https://huggingface.co/blog/llama2).
|
25 |
|
26 |
+
Buscando un modelo aún más potente? ¡Echa un vistazo a la demo del modelo grande [**70B**](https://huggingface.co/spaces/ysharma/Explore_llamav2_with_TGI)!
|
27 |
+
Para un modelo más pequeño que puedas ejecutar en muchas GPU, echa un vistazo a nuestra [demo del modelo 7B](https://huggingface.co/spaces/huggingface-projects/llama-2-7b-chat).
|
28 |
|
29 |
"""
|
30 |
|
|
|
32 |
<p/>
|
33 |
|
34 |
---
|
35 |
+
Como un trabajo derivado de [Llama-2-13b-chat](https://huggingface.co/meta-llama/Llama-2-13b-chat) de Meta,
|
36 |
+
esta demo está gobernada por la [licencia original](https://huggingface.co/spaces/huggingface-projects/llama-2-13b-chat/blob/main/LICENSE.txt) y la [política de uso aceptable](https://huggingface.co/spaces/huggingface-projects/llama-2-13b-chat/blob/main/USE_POLICY.md).
|
37 |
"""
|
38 |
|
39 |
if not torch.cuda.is_available():
|
40 |
+
DESCRIPTION += "\n<p>Running on CPU. This demo does not work on CPU.</p>"
|
41 |
|
42 |
|
43 |
+
model = None
|
44 |
+
tokenizer = None
|
45 |
+
|
46 |
+
@spaces.GPU
|
47 |
+
def load_model():
|
48 |
+
global model, tokenizer
|
49 |
model_id = "meta-llama/Llama-2-13b-chat-hf"
|
50 |
model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto", load_in_4bit=True)
|
51 |
tokenizer = AutoTokenizer.from_pretrained(model_id)
|
52 |
tokenizer.use_default_system_prompt = False
|
53 |
|
|
|
54 |
@spaces.GPU
|
55 |
def generate(
|
56 |
message: str,
|
|
|
96 |
yield "".join(outputs)
|
97 |
|
98 |
|
99 |
+
if gr.running:
|
100 |
+
load_model()
|
101 |
+
|
102 |
chat_interface = gr.ChatInterface(
|
103 |
fn=generate,
|
104 |
additional_inputs=[
|
|
|
139 |
value=1.2,
|
140 |
),
|
141 |
],
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
142 |
cache_examples=False,
|
143 |
)
|
144 |
|
test.py
ADDED
@@ -0,0 +1,32 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from transformers import AutoTokenizer, AutoModelForCausalLM
|
2 |
+
import os
|
3 |
+
import huggingface_hub
|
4 |
+
|
5 |
+
# Obtener token de Hugging Face
|
6 |
+
token = os.environ.get("HUGGINGFACE_HUB_TOKEN", None)
|
7 |
+
huggingface_hub.login(token=token)
|
8 |
+
|
9 |
+
# ID del modelo
|
10 |
+
model_id = "CohereForAI/aya-23-35B"
|
11 |
+
|
12 |
+
# Cargar tokenizador y modelo
|
13 |
+
tokenizer = AutoTokenizer.from_pretrained(model_id)
|
14 |
+
model = AutoModelForCausalLM.from_pretrained(model_id)
|
15 |
+
|
16 |
+
# Formatear mensaje con la plantilla de chat
|
17 |
+
messages = [{"role": "user", "content": "Anneme onu ne kadar sevdiğimi anlatan bir mektup yaz"}]
|
18 |
+
input_ids = tokenizer.apply_chat_template(messages, tokenize=True, add_generation_prompt=True, return_tensors="pt")
|
19 |
+
|
20 |
+
# Generar texto
|
21 |
+
gen_tokens = model.generate(
|
22 |
+
input_ids,
|
23 |
+
max_new_tokens=100,
|
24 |
+
do_sample=True,
|
25 |
+
temperature=0.3,
|
26 |
+
force_download=True
|
27 |
+
)
|
28 |
+
|
29 |
+
# Decodificar tokens generados
|
30 |
+
gen_text = tokenizer.decode(gen_tokens[0], skip_special_tokens=True)
|
31 |
+
print(gen_text)
|
32 |
+
|