Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
@@ -1,19 +1,188 @@
|
|
1 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
2 |
from transformers import AutoTokenizer, AutoModelForCausalLM
|
3 |
from peft import PeftModel
|
|
|
|
|
|
|
4 |
|
5 |
-
|
6 |
base_model = AutoModelForCausalLM.from_pretrained(
|
7 |
'meta-llama/Llama-2-7b-chat-hf',
|
8 |
trust_remote_code=True,
|
9 |
device_map="auto",
|
10 |
-
torch_dtype=torch.float16,
|
11 |
)
|
12 |
tokenizer = AutoTokenizer.from_pretrained('meta-llama/Llama-2-7b-chat-hf')
|
13 |
|
|
|
14 |
model = PeftModel.from_pretrained(base_model, 'FinGPT/fingpt-forecaster_dow30_llama2-7b_lora')
|
15 |
model = model.eval()
|
16 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
17 |
|
18 |
-
if __name__ == "
|
19 |
demo.queue(max_size=20).launch()
|
|
|
1 |
+
import os
|
2 |
+
from threading import Thread
|
3 |
+
from typing import Iterator, List, Tuple
|
4 |
+
|
5 |
+
import torch
|
6 |
+
from fastapi import FastAPI, HTTPException
|
7 |
+
from pydantic import BaseModel
|
8 |
from transformers import AutoTokenizer, AutoModelForCausalLM
|
9 |
from peft import PeftModel
|
10 |
+
import gradio as gr
|
11 |
+
from gradio import Blocks
|
12 |
+
from transformers import TextIteratorStreamer
|
13 |
|
14 |
+
# Load the base model and tokenizer
|
15 |
base_model = AutoModelForCausalLM.from_pretrained(
|
16 |
'meta-llama/Llama-2-7b-chat-hf',
|
17 |
trust_remote_code=True,
|
18 |
device_map="auto",
|
19 |
+
torch_dtype=torch.float16,
|
20 |
)
|
21 |
tokenizer = AutoTokenizer.from_pretrained('meta-llama/Llama-2-7b-chat-hf')
|
22 |
|
23 |
+
# Load the finetuned model
|
24 |
model = PeftModel.from_pretrained(base_model, 'FinGPT/fingpt-forecaster_dow30_llama2-7b_lora')
|
25 |
model = model.eval()
|
26 |
|
27 |
+
# Define constants
|
28 |
+
MAX_MAX_NEW_TOKENS = 2048
|
29 |
+
DEFAULT_MAX_NEW_TOKENS = 1024
|
30 |
+
MAX_INPUT_TOKEN_LENGTH = int(os.getenv("MAX_INPUT_TOKEN_LENGTH", "4096"))
|
31 |
+
|
32 |
+
# FastAPI setup
|
33 |
+
app = FastAPI()
|
34 |
+
|
35 |
+
class ChatRequest(BaseModel):
|
36 |
+
message: str
|
37 |
+
chat_history: List[Tuple[str, str]] = []
|
38 |
+
system_prompt: str = ""
|
39 |
+
max_new_tokens: int = 1024
|
40 |
+
temperature: float = 0.6
|
41 |
+
top_p: float = 0.9
|
42 |
+
top_k: int = 50
|
43 |
+
repetition_penalty: float = 1.2
|
44 |
+
|
45 |
+
@app.post("/chat/")
|
46 |
+
async def chat(request: ChatRequest):
|
47 |
+
try:
|
48 |
+
response = await generate_response(
|
49 |
+
request.message,
|
50 |
+
request.chat_history,
|
51 |
+
request.system_prompt,
|
52 |
+
request.max_new_tokens,
|
53 |
+
request.temperature,
|
54 |
+
request.top_p,
|
55 |
+
request.top_k,
|
56 |
+
request.repetition_penalty
|
57 |
+
)
|
58 |
+
return {"response": response}
|
59 |
+
except Exception as e:
|
60 |
+
raise HTTPException(status_code=500, detail=str(e))
|
61 |
+
|
62 |
+
async def generate_response(
|
63 |
+
message: str,
|
64 |
+
chat_history: List[Tuple[str, str]],
|
65 |
+
system_prompt: str,
|
66 |
+
max_new_tokens: int = 1024,
|
67 |
+
temperature: float = 0.6,
|
68 |
+
top_p: float = 0.9,
|
69 |
+
top_k: int = 50,
|
70 |
+
repetition_penalty: float = 1.2,
|
71 |
+
) -> str:
|
72 |
+
conversation = []
|
73 |
+
if system_prompt:
|
74 |
+
conversation.append({"role": "system", "content": system_prompt})
|
75 |
+
for user, assistant in chat_history:
|
76 |
+
conversation.extend([{"role": "user", "content": user}, {"role": "assistant", "content": assistant}])
|
77 |
+
conversation.append({"role": "user", "content": message})
|
78 |
+
|
79 |
+
input_ids = tokenizer.apply_chat_template(conversation, return_tensors="pt")
|
80 |
+
if input_ids.shape[1] > MAX_INPUT_TOKEN_LENGTH:
|
81 |
+
input_ids = input_ids[:, -MAX_INPUT_TOKEN_LENGTH:]
|
82 |
+
|
83 |
+
input_ids = input_ids.to(model.device)
|
84 |
+
|
85 |
+
streamer = TextIteratorStreamer(tokenizer, timeout=10.0, skip_prompt=True, skip_special_tokens=True)
|
86 |
+
generate_kwargs = {
|
87 |
+
"input_ids": input_ids,
|
88 |
+
"streamer": streamer,
|
89 |
+
"max_new_tokens": max_new_tokens,
|
90 |
+
"do_sample": True,
|
91 |
+
"top_p": top_p,
|
92 |
+
"top_k": top_k,
|
93 |
+
"temperature": temperature,
|
94 |
+
"num_beams": 1,
|
95 |
+
"repetition_penalty": repetition_penalty,
|
96 |
+
}
|
97 |
+
t = Thread(target=model.generate, kwargs=generate_kwargs)
|
98 |
+
t.start()
|
99 |
+
|
100 |
+
outputs = []
|
101 |
+
for text in streamer:
|
102 |
+
outputs.append(text)
|
103 |
+
return "".join(outputs)
|
104 |
+
|
105 |
+
# Gradio setup
|
106 |
+
def generate(
|
107 |
+
message: str,
|
108 |
+
chat_history: List[Tuple[str, str]],
|
109 |
+
system_prompt: str,
|
110 |
+
max_new_tokens: int = 1024,
|
111 |
+
temperature: float = 0.6,
|
112 |
+
top_p: float = 0.9,
|
113 |
+
top_k: int = 50,
|
114 |
+
repetition_penalty: float = 1.2,
|
115 |
+
) -> Iterator[str]:
|
116 |
+
return generate_response(
|
117 |
+
message,
|
118 |
+
chat_history,
|
119 |
+
system_prompt,
|
120 |
+
max_new_tokens,
|
121 |
+
temperature,
|
122 |
+
top_p,
|
123 |
+
top_k,
|
124 |
+
repetition_penalty
|
125 |
+
)
|
126 |
+
|
127 |
+
chat_interface = gr.ChatInterface(
|
128 |
+
fn=generate,
|
129 |
+
additional_inputs=[
|
130 |
+
gr.Textbox(label="System prompt", lines=6),
|
131 |
+
gr.Slider(
|
132 |
+
label="Max new tokens",
|
133 |
+
minimum=1,
|
134 |
+
maximum=MAX_MAX_NEW_TOKENS,
|
135 |
+
step=1,
|
136 |
+
value=DEFAULT_MAX_NEW_TOKENS,
|
137 |
+
),
|
138 |
+
gr.Slider(
|
139 |
+
label="Temperature",
|
140 |
+
minimum=0.1,
|
141 |
+
maximum=4.0,
|
142 |
+
step=0.1,
|
143 |
+
value=0.6,
|
144 |
+
),
|
145 |
+
gr.Slider(
|
146 |
+
label="Top-p (nucleus sampling)",
|
147 |
+
minimum=0.05,
|
148 |
+
maximum=1.0,
|
149 |
+
step=0.05,
|
150 |
+
value=0.9,
|
151 |
+
),
|
152 |
+
gr.Slider(
|
153 |
+
label="Top-k",
|
154 |
+
minimum=1,
|
155 |
+
maximum=1000,
|
156 |
+
step=1,
|
157 |
+
value=50,
|
158 |
+
),
|
159 |
+
gr.Slider(
|
160 |
+
label="Repetition penalty",
|
161 |
+
minimum=1.0,
|
162 |
+
maximum=2.0,
|
163 |
+
step=0.05,
|
164 |
+
value=1.2,
|
165 |
+
),
|
166 |
+
],
|
167 |
+
stop_btn=None,
|
168 |
+
examples=[
|
169 |
+
["Hello there! How are you doing?"],
|
170 |
+
["Can you explain briefly to me what is the Python programming language?"],
|
171 |
+
["Explain the plot of Cinderella in a sentence."],
|
172 |
+
["How many hours does it take a man to eat a Helicopter?"],
|
173 |
+
["Write a 100-word article on 'Benefits of Open-Source in AI research'"],
|
174 |
+
],
|
175 |
+
)
|
176 |
+
|
177 |
+
with Blocks(css="style.css") as demo:
|
178 |
+
gr.Markdown("# Llama-2 7B Chat")
|
179 |
+
gr.Markdown("""
|
180 |
+
This Space demonstrates the Llama-2 7B Chat model by Meta, fine-tuned for chat instructions.
|
181 |
+
Feel free to chat with the model here or use the API to integrate it into your applications.
|
182 |
+
""")
|
183 |
+
chat_interface.render()
|
184 |
+
gr.Markdown("---")
|
185 |
+
gr.Markdown("This demo is governed by the original [license](https://huggingface.co/meta-llama/Llama-2-7b-chat-hf/blob/main/LICENSE.txt).")
|
186 |
|
187 |
+
if __name__ == "__main__":
|
188 |
demo.queue(max_size=20).launch()
|