Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
@@ -1,5 +1,5 @@
|
|
1 |
from threading import Thread
|
2 |
-
|
3 |
import torch
|
4 |
import gradio as gr
|
5 |
import re
|
@@ -8,56 +8,18 @@ torch_device = "cuda" if torch.cuda.is_available() else "cpu"
|
|
8 |
print("Running on device:", torch_device)
|
9 |
print("CPU threads:", torch.get_num_threads())
|
10 |
|
11 |
-
|
12 |
-
|
13 |
-
|
14 |
-
# 20k or 30k
|
15 |
-
#18k > 상대의 말까지 하는 이슈가 있음
|
16 |
-
#8k > 약간 아쉬운가?
|
17 |
-
|
18 |
-
base_model = AutoModelForCausalLM.from_pretrained("Skyranch/KoAlpaca-Polyglot-12.8B-ggml-model-f16", hf=True, model_type='gpt_neox')
|
19 |
-
tokenizer = AutoTokenizer.from_pretrained("beomi/KoAlpaca-Polyglot-12.8B")
|
20 |
-
|
21 |
-
#base_model = AutoModelForCausalLM.from_pretrained("EleutherAI/polyglot-ko-3.8b")
|
22 |
-
#tokenizer = AutoTokenizer.from_pretrained("EleutherAI/polyglot-ko-3.8b")
|
23 |
-
base_model.eval()
|
24 |
-
base_model.config.use_cache = True
|
25 |
-
|
26 |
-
|
27 |
-
#model = PeftModel.from_pretrained(base_model, peft_model_id, device_map="auto")
|
28 |
-
#model.eval()
|
29 |
-
#model.config.use_cache = True
|
30 |
-
|
31 |
-
def gen(x, top_p, top_k, temperature, max_new_tokens, repetition_penalty):
|
32 |
-
gened = base_model.generate(
|
33 |
-
**tokenizer(
|
34 |
-
f"{x}",
|
35 |
-
return_tensors='pt',
|
36 |
-
return_token_type_ids=False
|
37 |
-
),
|
38 |
-
#bad_words_ids = bad_words_ids ,
|
39 |
-
max_new_tokens=max_new_tokens,
|
40 |
-
min_new_tokens = 5,
|
41 |
-
exponential_decay_length_penalty = (max_new_tokens/2, 1.1),
|
42 |
-
top_p=top_p,
|
43 |
-
top_k=top_k,
|
44 |
-
temperature = temperature,
|
45 |
-
early_stopping=True,
|
46 |
-
do_sample=True,
|
47 |
-
eos_token_id=2,
|
48 |
-
pad_token_id=2,
|
49 |
-
#stopping_criteria = stopping_criteria,
|
50 |
-
repetition_penalty=repetition_penalty,
|
51 |
-
no_repeat_ngram_size = 2
|
52 |
-
)
|
53 |
|
54 |
-
|
55 |
-
|
|
|
|
|
56 |
|
57 |
def reset_textbox():
|
58 |
return gr.update(value='')
|
59 |
|
60 |
-
|
61 |
with gr.Blocks() as demo:
|
62 |
duplicate_link = "https://huggingface.co/spaces/beomi/KoRWKV-1.5B?duplicate=true"
|
63 |
gr.Markdown(
|
@@ -67,7 +29,7 @@ with gr.Blocks() as demo:
|
|
67 |
with gr.Row():
|
68 |
with gr.Column(scale=4):
|
69 |
user_text = gr.Textbox(
|
70 |
-
placeholder='
|
71 |
label="User input"
|
72 |
)
|
73 |
model_output = gr.Textbox(label="Model output", lines=10, interactive=False)
|
@@ -76,18 +38,6 @@ with gr.Blocks() as demo:
|
|
76 |
max_new_tokens = gr.Slider(
|
77 |
minimum=1, maximum=200, value=20, step=1, interactive=True, label="Max New Tokens",
|
78 |
)
|
79 |
-
top_p = gr.Slider(
|
80 |
-
minimum=0.05, maximum=1.0, value=0.8, step=0.05, interactive=True, label="Top-p (nucleus sampling)",
|
81 |
-
)
|
82 |
-
top_k = gr.Slider(
|
83 |
-
minimum=5, maximum=100, value=30, step=5, interactive=True, label="Top-k (nucleus sampling)",
|
84 |
-
)
|
85 |
-
temperature = gr.Slider(
|
86 |
-
minimum=0.1, maximum=2.0, value=0.5, step=0.1, interactive=True, label="Temperature",
|
87 |
-
)
|
88 |
-
repetition_penalty = gr.Slider(
|
89 |
-
minimum=1.0, maximum=3.0, value=1.2, step=0.1, interactive=True, label="repetition_penalty",
|
90 |
-
)
|
91 |
|
92 |
-
button_submit.click(gen, [user_text,
|
93 |
demo.queue(max_size=32).launch(enable_queue=True)
|
|
|
1 |
from threading import Thread
|
2 |
+
from llama_cpp import Llama
|
3 |
import torch
|
4 |
import gradio as gr
|
5 |
import re
|
|
|
8 |
print("Running on device:", torch_device)
|
9 |
print("CPU threads:", torch.get_num_threads())
|
10 |
|
11 |
+
llm = Llama(model_path = 'Llama-2-ko-7B-chat-gguf-q4_0.bin',
|
12 |
+
n_ctx=1024,
|
13 |
+
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
14 |
|
15 |
+
def gen(x, max_new_tokens):
|
16 |
+
output = llm(f"Q: {x} A: ", max_tokens=1024, stop=["Q:", "\n"], echo=True)
|
17 |
+
|
18 |
+
return output['choices'][0]['text'].replace('▁',' ')
|
19 |
|
20 |
def reset_textbox():
|
21 |
return gr.update(value='')
|
22 |
|
|
|
23 |
with gr.Blocks() as demo:
|
24 |
duplicate_link = "https://huggingface.co/spaces/beomi/KoRWKV-1.5B?duplicate=true"
|
25 |
gr.Markdown(
|
|
|
29 |
with gr.Row():
|
30 |
with gr.Column(scale=4):
|
31 |
user_text = gr.Textbox(
|
32 |
+
placeholder='우리 여행 갈래?',
|
33 |
label="User input"
|
34 |
)
|
35 |
model_output = gr.Textbox(label="Model output", lines=10, interactive=False)
|
|
|
38 |
max_new_tokens = gr.Slider(
|
39 |
minimum=1, maximum=200, value=20, step=1, interactive=True, label="Max New Tokens",
|
40 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
41 |
|
42 |
+
button_submit.click(gen, [user_text, max_new_tokens], model_output)
|
43 |
demo.queue(max_size=32).launch(enable_queue=True)
|