Spaces:
Runtime error
Runtime error
typesdigital
commited on
Commit
•
cba212b
1
Parent(s):
3d4072e
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,56 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
from peft import PeftModel, PeftConfig
|
3 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig, AutoTokenizer, GenerationConfig
|
4 |
+
|
5 |
+
peft_model_id = "mrm8488/falcon-7b-ft-codeAlpaca_20k-v2" # adapter
|
6 |
+
config = PeftConfig.from_pretrained(peft_model_id)
|
7 |
+
model = AutoModelForCausalLM.from_pretrained(config.base_model_name_or_path, return_dict=True, load_in_8bit=True, device_map={"":0}, trust_remote_code=True)
|
8 |
+
tokenizer = AutoTokenizer.from_pretrained(peft_model_id)
|
9 |
+
|
10 |
+
model = PeftModel.from_pretrained(model, peft_model_id)
|
11 |
+
model.eval()
|
12 |
+
|
13 |
+
def generate(
|
14 |
+
instruction,
|
15 |
+
max_new_tokens=128,
|
16 |
+
temperature=0.1,
|
17 |
+
top_p=0.75,
|
18 |
+
top_k=40,
|
19 |
+
num_beams=4,
|
20 |
+
**kwargs
|
21 |
+
):
|
22 |
+
prompt = instruction + "\n### Solution:\n"
|
23 |
+
print(prompt)
|
24 |
+
inputs = tokenizer(prompt, return_tensors="pt")
|
25 |
+
input_ids = inputs["input_ids"].to("cuda")
|
26 |
+
attention_mask = inputs["attention_mask"].to("cuda")
|
27 |
+
generation_config = GenerationConfig(
|
28 |
+
temperature=temperature,
|
29 |
+
top_p=top_p,
|
30 |
+
top_k=top_k,
|
31 |
+
num_beams=num_beams,
|
32 |
+
**kwargs,
|
33 |
+
)
|
34 |
+
with torch.no_grad():
|
35 |
+
generation_output = model.generate(
|
36 |
+
input_ids=input_ids,
|
37 |
+
attention_mask=attention_mask,
|
38 |
+
generation_config=generation_config,
|
39 |
+
return_dict_in_generate=True,
|
40 |
+
output_scores=True,
|
41 |
+
max_new_tokens=max_new_tokens,
|
42 |
+
early_stopping=True
|
43 |
+
)
|
44 |
+
s = generation_output.sequences[0]
|
45 |
+
output = tokenizer.decode(s)
|
46 |
+
return output.split("### Solution:")[1].lstrip("\n")
|
47 |
+
|
48 |
+
import gradio as gr
|
49 |
+
def my_function(input):
|
50 |
+
# Perform your task or computation using the input
|
51 |
+
# Return the output/result
|
52 |
+
return output
|
53 |
+
|
54 |
+
iface = gr.Interface(fn=my_function, inputs="text", outputs="text")
|
55 |
+
iface.launch()
|
56 |
+
|