Spaces:
Sleeping
Sleeping
Commit
·
6c242fa
1
Parent(s):
01db43e
yee
Browse files
gpt.py
CHANGED
|
@@ -113,7 +113,7 @@ torch.set_float32_matmul_precision('high')
|
|
| 113 |
my_GPT = GPT(enc.n_vocab, block_size, n_layers, n_heads, d_model, dropout=0.1) #enc.n_vocab
|
| 114 |
my_GPT = my_GPT.to(device)
|
| 115 |
my_GPT = torch.compile(my_GPT)
|
| 116 |
-
my_GPT.load_state_dict(torch.load('latest_model_finetune.pth'))
|
| 117 |
my_GPT.eval()
|
| 118 |
|
| 119 |
eot = enc._special_tokens['<|endoftext|>']
|
|
|
|
| 113 |
my_GPT = GPT(enc.n_vocab, block_size, n_layers, n_heads, d_model, dropout=0.1) #enc.n_vocab
|
| 114 |
my_GPT = my_GPT.to(device)
|
| 115 |
my_GPT = torch.compile(my_GPT)
|
| 116 |
+
my_GPT.load_state_dict(torch.load('latest_model_finetune.pth', map_location=torch.device('cpu')))
|
| 117 |
my_GPT.eval()
|
| 118 |
|
| 119 |
eot = enc._special_tokens['<|endoftext|>']
|