Update app.py
Browse files
app.py
CHANGED
@@ -1,8 +1,25 @@
|
|
1 |
import gradio as gr
|
2 |
import torch
|
3 |
-
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
4 |
|
5 |
-
model = AutoModelForCausalLM.from_pretrained("checkpoint_500",trust_remote_code=True)
|
6 |
tokenizer = AutoTokenizer.from_pretrained("checkpoint_500", trust_remote_code=True)
|
7 |
tokenizer.pad_token = tokenizer.eos_token
|
8 |
|
|
|
1 |
import gradio as gr
|
2 |
import torch
|
3 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline,BitsAndBytesConfig
|
4 |
+
|
5 |
+
#model = AutoModelForCausalLM.from_pretrained("checkpoint_500",trust_remote_code=True)
|
6 |
+
|
7 |
+
model_name = "microsoft/phi-2"
|
8 |
+
|
9 |
+
bnb_config = BitsAndBytesConfig(
|
10 |
+
load_in_4bit=True,
|
11 |
+
bnb_4bit_quant_type="nf4",
|
12 |
+
bnb_4bit_compute_dtype=torch.float16,
|
13 |
+
)
|
14 |
+
|
15 |
+
model = AutoModelForCausalLM.from_pretrained(
|
16 |
+
model_name,
|
17 |
+
quantization_config=bnb_config,
|
18 |
+
trust_remote_code=True
|
19 |
+
)
|
20 |
+
model.config.use_cache = False
|
21 |
+
model.load_adapter("checkpoint")
|
22 |
|
|
|
23 |
tokenizer = AutoTokenizer.from_pretrained("checkpoint_500", trust_remote_code=True)
|
24 |
tokenizer.pad_token = tokenizer.eos_token
|
25 |
|