Upload app.py
Browse files
app.py
ADDED
@@ -0,0 +1,141 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
from peft import PeftModel
|
3 |
+
import transformers
|
4 |
+
import gradio as gr
|
5 |
+
|
6 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer, GenerationConfig
|
7 |
+
|
8 |
+
tokenizer = AutoTokenizer.from_pretrained("mistralai/Mistral-7B-v0.1")
|
9 |
+
|
10 |
+
BASE_MODEL = "mistralai/Mistral-7B-v0.1"
|
11 |
+
LORA_WEIGHTS = "./qlora-out.mistral.0.9978/"
|
12 |
+
|
13 |
+
if torch.cuda.is_available():
|
14 |
+
device = "cuda"
|
15 |
+
else:
|
16 |
+
device = "cpu"
|
17 |
+
|
18 |
+
try:
|
19 |
+
if torch.backends.mps.is_available():
|
20 |
+
device = "mps"
|
21 |
+
except:
|
22 |
+
pass
|
23 |
+
|
24 |
+
if device == "cuda":
|
25 |
+
from transformers import BitsAndBytesConfig
|
26 |
+
|
27 |
+
nf4_config = BitsAndBytesConfig(
|
28 |
+
load_in_4bit=True,
|
29 |
+
bnb_4bit_quant_type="nf4",
|
30 |
+
bnb_4bit_use_double_quant=True,
|
31 |
+
bnb_4bit_compute_dtype=torch.bfloat16
|
32 |
+
)
|
33 |
+
|
34 |
+
model = AutoModelForCausalLM.from_pretrained(BASE_MODEL, quantization_config=nf4_config)
|
35 |
+
|
36 |
+
model = PeftModel.from_pretrained(
|
37 |
+
model, LORA_WEIGHTS, torch_dtype=torch.float16, force_download=True
|
38 |
+
)
|
39 |
+
elif device == "mps":
|
40 |
+
model = AutoModelForCausalLM.from_pretrained(
|
41 |
+
BASE_MODEL,
|
42 |
+
device_map={"": device},
|
43 |
+
torch_dtype=torch.float16,
|
44 |
+
)
|
45 |
+
model = PeftModel.from_pretrained(
|
46 |
+
model,
|
47 |
+
LORA_WEIGHTS,
|
48 |
+
device_map={"": device},
|
49 |
+
torch_dtype=torch.float16,
|
50 |
+
)
|
51 |
+
else:
|
52 |
+
model = AutoModelForCausalLM.from_pretrained(
|
53 |
+
BASE_MODEL, device_map={"": device}, low_cpu_mem_usage=True
|
54 |
+
)
|
55 |
+
model = PeftModel.from_pretrained(
|
56 |
+
model,
|
57 |
+
LORA_WEIGHTS,
|
58 |
+
device_map={"": device},
|
59 |
+
)
|
60 |
+
|
61 |
+
|
62 |
+
def generate_prompt(instruction, input=None):
|
63 |
+
if input:
|
64 |
+
return f"""Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.
|
65 |
+
|
66 |
+
### Instruction:
|
67 |
+
{instruction}
|
68 |
+
|
69 |
+
### Input:
|
70 |
+
{input}
|
71 |
+
|
72 |
+
### Response:"""
|
73 |
+
else:
|
74 |
+
return f"""Below is an instruction that describes a task. Write a response that appropriately completes the request.
|
75 |
+
|
76 |
+
### Instruction:
|
77 |
+
{instruction}
|
78 |
+
|
79 |
+
### Response:"""
|
80 |
+
|
81 |
+
if device != "cpu":
|
82 |
+
model.half()
|
83 |
+
model.eval()
|
84 |
+
if torch.__version__ >= "2":
|
85 |
+
model = torch.compile(model)
|
86 |
+
|
87 |
+
|
88 |
+
def evaluate(
|
89 |
+
instruction,
|
90 |
+
input=None,
|
91 |
+
temperature=0.1,
|
92 |
+
top_p=0.75,
|
93 |
+
top_k=40,
|
94 |
+
num_beams=4,
|
95 |
+
max_new_tokens=128,
|
96 |
+
**kwargs,
|
97 |
+
):
|
98 |
+
prompt = generate_prompt(instruction, input)
|
99 |
+
inputs = tokenizer(prompt, return_tensors="pt")
|
100 |
+
input_ids = inputs["input_ids"].to(device)
|
101 |
+
generation_config = GenerationConfig(
|
102 |
+
temperature=temperature,
|
103 |
+
top_p=top_p,
|
104 |
+
top_k=top_k,
|
105 |
+
num_beams=num_beams,
|
106 |
+
**kwargs,
|
107 |
+
)
|
108 |
+
with torch.no_grad():
|
109 |
+
generation_output = model.generate(
|
110 |
+
input_ids=input_ids,
|
111 |
+
generation_config=generation_config,
|
112 |
+
return_dict_in_generate=True,
|
113 |
+
output_scores=True,
|
114 |
+
max_new_tokens=max_new_tokens,
|
115 |
+
)
|
116 |
+
s = generation_output.sequences[0]
|
117 |
+
output = tokenizer.decode(s)
|
118 |
+
return output.split("### Response:")[1].strip()
|
119 |
+
|
120 |
+
|
121 |
+
g = gr.Interface(
|
122 |
+
fn=evaluate,
|
123 |
+
inputs=[
|
124 |
+
gr.components.Textbox(
|
125 |
+
lines=2, label="Utasítás", placeholder="Mesélj kicsit a szürkemarháról!"
|
126 |
+
),
|
127 |
+
gr.components.Textbox(lines=2, label="Input", placeholder="üres"),
|
128 |
+
gr.components.Slider(minimum=0, maximum=1, value=0.1, label="Temperature"),
|
129 |
+
gr.components.Slider(minimum=0, maximum=1, value=0.75, label="Top p"),
|
130 |
+
gr.components.Slider(minimum=0, maximum=100, step=1, value=40, label="Top k"),
|
131 |
+
gr.components.Slider(minimum=1, maximum=4, step=1, value=4, label="Beams"),
|
132 |
+
gr.components.Slider(
|
133 |
+
minimum=1, maximum=512, step=1, value=128, label="Max tokens"
|
134 |
+
),
|
135 |
+
],
|
136 |
+
outputs=["text"],
|
137 |
+
title="szürkemarha-mistral-v1",
|
138 |
+
description="A szürkemarha-mistral egy fejlesztés alatt álló 7 milliárd paraméteres Mistral-0.1 alapú model LoRA finomhangolva instrukciókövetésre.",
|
139 |
+
)
|
140 |
+
g.queue(concurrency_count=1)
|
141 |
+
g.launch()
|