Spaces:
Sleeping
Sleeping
Commit
·
18d159f
1
Parent(s):
a802987
changed for error libcuda.so
Browse files
app.py
CHANGED
@@ -9,31 +9,6 @@ dataset_name = "timdettmers/openassistant-guanaco"
|
|
9 |
dataset = load_dataset(dataset_name, split="train")
|
10 |
|
11 |
|
12 |
-
|
13 |
-
|
14 |
-
# quantizition configuration
|
15 |
-
bnb_config = BitsAndBytesConfig(
|
16 |
-
load_in_4bit=True,
|
17 |
-
bnb_4bit_quant_type="nf4",
|
18 |
-
bnb_4bit_compute_dtype=torch.float16,
|
19 |
-
)
|
20 |
-
|
21 |
-
# download model
|
22 |
-
|
23 |
-
model_name = "TinyPixel/Llama-2-7B-bf16-sharded"
|
24 |
-
model = AutoModelForCausalLM.from_pretrained(
|
25 |
-
model_name,
|
26 |
-
quantization_config=bnb_config,
|
27 |
-
trust_remote_code=True
|
28 |
-
)
|
29 |
-
model.config.use_cache = False
|
30 |
-
|
31 |
-
tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
|
32 |
-
tokenizer.pad_token = tokenizer.eos_token
|
33 |
-
|
34 |
-
text = "What is a large language model?"
|
35 |
-
device = "cuda:0"
|
36 |
-
inputs = tokenizer(text, return_tensors="pt").to(device)
|
37 |
-
outputs = model.generate(**inputs, max_new_tokens=50)
|
38 |
-
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
|
39 |
|
|
|
9 |
dataset = load_dataset(dataset_name, split="train")
|
10 |
|
11 |
|
12 |
+
echo $LD_LIBRARY_PATH #path
|
13 |
+
sudo find /usr/ -name 'libcuda.so.*' #version
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
14 |
|