Update app.py
Browse files
app.py
CHANGED
@@ -2,64 +2,49 @@ import torch
|
|
2 |
from PIL import Image
|
3 |
import gradio as gr
|
4 |
import spaces
|
5 |
-
from transformers import AutoModelForCausalLM, AutoTokenizer,
|
6 |
import os
|
7 |
from threading import Thread
|
8 |
|
9 |
|
10 |
HF_TOKEN = os.environ.get("HF_TOKEN", None)
|
11 |
-
MODEL_ID = "
|
12 |
-
MODEL_ID2 = "CohereForAI/aya-23-35B"
|
13 |
MODELS = os.environ.get("MODELS")
|
14 |
MODEL_NAME = MODELS.split("/")[-1]
|
15 |
|
16 |
-
TITLE = "<h1><center>
|
17 |
|
18 |
-
DESCRIPTION = f
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
19 |
|
20 |
CSS = """
|
21 |
.duplicate-button {
|
22 |
-
|
23 |
-
|
24 |
-
|
25 |
-
|
|
|
|
|
|
|
26 |
}
|
27 |
"""
|
28 |
|
29 |
-
|
30 |
-
#QUANTIZE
|
31 |
-
QUANTIZE_4BIT = True
|
32 |
-
USE_GRAD_CHECKPOINTING = True
|
33 |
-
TRAIN_BATCH_SIZE = 2
|
34 |
-
TRAIN_MAX_SEQ_LENGTH = 512
|
35 |
-
USE_FLASH_ATTENTION = False
|
36 |
-
GRAD_ACC_STEPS = 16
|
37 |
-
|
38 |
-
quantization_config = None
|
39 |
-
|
40 |
-
if QUANTIZE_4BIT:
|
41 |
-
quantization_config = BitsAndBytesConfig(
|
42 |
-
load_in_4bit=True,
|
43 |
-
bnb_4bit_quant_type="nf4",
|
44 |
-
bnb_4bit_use_double_quant=True,
|
45 |
-
bnb_4bit_compute_dtype=torch.bfloat16,
|
46 |
-
)
|
47 |
-
|
48 |
-
attn_implementation = None
|
49 |
-
if USE_FLASH_ATTENTION:
|
50 |
-
attn_implementation="flash_attention_2"
|
51 |
-
|
52 |
model = AutoModelForCausalLM.from_pretrained(
|
53 |
MODELS,
|
54 |
-
|
55 |
-
attn_implementation=attn_implementation,
|
56 |
-
torch_dtype=torch.bfloat16,
|
57 |
device_map="auto",
|
58 |
)
|
59 |
tokenizer = AutoTokenizer.from_pretrained(MODELS)
|
60 |
|
61 |
@spaces.GPU
|
62 |
-
def stream_chat(message: str, history: list, temperature: float, max_new_tokens: int):
|
63 |
print(f'message is - {message}')
|
64 |
print(f'history is - {history}')
|
65 |
conversation = []
|
@@ -69,16 +54,21 @@ def stream_chat(message: str, history: list, temperature: float, max_new_tokens:
|
|
69 |
|
70 |
print(f"Conversation is -\n{conversation}")
|
71 |
|
72 |
-
input_ids = tokenizer.apply_chat_template(conversation, tokenize=
|
|
|
73 |
|
74 |
-
streamer = TextIteratorStreamer(tokenizer,
|
75 |
|
76 |
generate_kwargs = dict(
|
77 |
-
|
78 |
streamer=streamer,
|
|
|
|
|
|
|
79 |
max_new_tokens=max_new_tokens,
|
80 |
do_sample=True,
|
81 |
temperature=temperature,
|
|
|
82 |
)
|
83 |
|
84 |
thread = Thread(target=model.generate, kwargs=generate_kwargs)
|
@@ -119,6 +109,30 @@ with gr.Blocks(css=CSS) as demo:
|
|
119 |
label="Max new tokens",
|
120 |
render=False,
|
121 |
),
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
122 |
],
|
123 |
examples=[
|
124 |
["Help me study vocabulary: write a sentence for me to fill in the blank, and I'll try to pick the correct option."],
|
|
|
2 |
from PIL import Image
|
3 |
import gradio as gr
|
4 |
import spaces
|
5 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
|
6 |
import os
|
7 |
from threading import Thread
|
8 |
|
9 |
|
10 |
HF_TOKEN = os.environ.get("HF_TOKEN", None)
|
11 |
+
MODEL_ID = "Qwen/Qwen1.5-7B-Chat"
|
|
|
12 |
MODELS = os.environ.get("MODELS")
|
13 |
MODEL_NAME = MODELS.split("/")[-1]
|
14 |
|
15 |
+
TITLE = "<h1><center>Qwen2-Chatbox</center></h1>"
|
16 |
|
17 |
+
DESCRIPTION = f"""
|
18 |
+
<h3>MODEL: <a href="https://hf.co/{MODELS}">{MODEL_NAME}</a></h3>
|
19 |
+
<center>
|
20 |
+
<p>Qwen is the large language model built by Alibaba Cloud.
|
21 |
+
<br>
|
22 |
+
Feel free to test without log.
|
23 |
+
</p>
|
24 |
+
</center>
|
25 |
+
"""
|
26 |
|
27 |
CSS = """
|
28 |
.duplicate-button {
|
29 |
+
margin: auto !important;
|
30 |
+
color: white !important;
|
31 |
+
background: black !important;
|
32 |
+
border-radius: 100vh !important;
|
33 |
+
}
|
34 |
+
h3 {
|
35 |
+
text-align: center;
|
36 |
}
|
37 |
"""
|
38 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
39 |
model = AutoModelForCausalLM.from_pretrained(
|
40 |
MODELS,
|
41 |
+
torch_dtype=torch.float16,
|
|
|
|
|
42 |
device_map="auto",
|
43 |
)
|
44 |
tokenizer = AutoTokenizer.from_pretrained(MODELS)
|
45 |
|
46 |
@spaces.GPU
|
47 |
+
def stream_chat(message: str, history: list, temperature: float, max_new_tokens: int, top_p: float, top_k: int, penalty: float):
|
48 |
print(f'message is - {message}')
|
49 |
print(f'history is - {history}')
|
50 |
conversation = []
|
|
|
54 |
|
55 |
print(f"Conversation is -\n{conversation}")
|
56 |
|
57 |
+
input_ids = tokenizer.apply_chat_template(conversation, tokenize=False, add_generation_prompt=True)
|
58 |
+
inputs = tokenizer(input_ids, return_tensors="pt").to(0)
|
59 |
|
60 |
+
streamer = TextIteratorStreamer(tokenizer, timeout=10., skip_prompt=True, skip_special_tokens=True)
|
61 |
|
62 |
generate_kwargs = dict(
|
63 |
+
inputs,
|
64 |
streamer=streamer,
|
65 |
+
top_k=top_k,
|
66 |
+
top_p=top_p,
|
67 |
+
repetition_penalty=penalty,
|
68 |
max_new_tokens=max_new_tokens,
|
69 |
do_sample=True,
|
70 |
temperature=temperature,
|
71 |
+
eos_token_id = [151645, 151643],
|
72 |
)
|
73 |
|
74 |
thread = Thread(target=model.generate, kwargs=generate_kwargs)
|
|
|
109 |
label="Max new tokens",
|
110 |
render=False,
|
111 |
),
|
112 |
+
gr.Slider(
|
113 |
+
minimum=0.0,
|
114 |
+
maximum=1.0,
|
115 |
+
step=0.1,
|
116 |
+
value=0.8,
|
117 |
+
label="top_p",
|
118 |
+
render=False,
|
119 |
+
),
|
120 |
+
gr.Slider(
|
121 |
+
minimum=1,
|
122 |
+
maximum=20,
|
123 |
+
step=1,
|
124 |
+
value=20,
|
125 |
+
label="top_k",
|
126 |
+
render=False,
|
127 |
+
),
|
128 |
+
gr.Slider(
|
129 |
+
minimum=0.0,
|
130 |
+
maximum=2.0,
|
131 |
+
step=0.1,
|
132 |
+
value=1.0,
|
133 |
+
label="Repetition penalty",
|
134 |
+
render=False,
|
135 |
+
),
|
136 |
],
|
137 |
examples=[
|
138 |
["Help me study vocabulary: write a sentence for me to fill in the blank, and I'll try to pick the correct option."],
|