Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -3,19 +3,16 @@ import torch.nn as nn
|
|
3 |
from torch.nn import functional as F
|
4 |
import tiktoken
|
5 |
import gradio as gr
|
6 |
-
import asyncio
|
7 |
|
8 |
-
# Try to import spaces, use a dummy decorator if not available
|
9 |
try:
|
10 |
import spaces
|
11 |
use_spaces_gpu = True
|
12 |
except ImportError:
|
13 |
use_spaces_gpu = False
|
14 |
-
# Dummy decorator in case spaces is not available
|
15 |
def dummy_gpu_decorator(func):
|
16 |
return func
|
17 |
spaces = type('', (), {'GPU': dummy_gpu_decorator})()
|
18 |
-
|
19 |
# Define the GPTConfig class
|
20 |
class GPTConfig:
|
21 |
def __init__(self):
|
@@ -131,10 +128,10 @@ def load_model(model_path):
|
|
131 |
|
132 |
enc = tiktoken.get_encoding('gpt2')
|
133 |
|
|
|
134 |
# Update the generate_text function
|
135 |
-
@spaces.GPU
|
136 |
-
|
137 |
-
# Load the model inside the GPU-decorated function
|
138 |
model = load_model('gpt_model.pth')
|
139 |
device = next(model.parameters()).device
|
140 |
input_ids = torch.tensor(enc.encode(prompt)).unsqueeze(0).to(device)
|
@@ -153,24 +150,15 @@ async def generate_text(prompt, max_length=432, temperature=0.8, top_k=40):
|
|
153 |
input_ids = torch.cat([input_ids, next_token], dim=-1)
|
154 |
generated.append(next_token.item())
|
155 |
|
156 |
-
next_token_str = enc.decode([next_token.item()])
|
157 |
-
yield next_token_str
|
158 |
-
|
159 |
if next_token.item() == enc.encode('\n')[0] and len(generated) > 100:
|
160 |
break
|
161 |
-
|
162 |
-
await asyncio.sleep(0.02)
|
163 |
|
164 |
-
|
165 |
-
yield "... (output truncated due to length)"
|
166 |
|
167 |
# Add the gradio_generate function
|
168 |
-
@spaces.GPU
|
169 |
-
|
170 |
-
|
171 |
-
async for token in generate_text(prompt, max_length, temperature, top_k):
|
172 |
-
output += token
|
173 |
-
yield output
|
174 |
|
175 |
|
176 |
# # Your existing imports and model code here...
|
|
|
3 |
from torch.nn import functional as F
|
4 |
import tiktoken
|
5 |
import gradio as gr
|
|
|
6 |
|
|
|
7 |
try:
|
8 |
import spaces
|
9 |
use_spaces_gpu = True
|
10 |
except ImportError:
|
11 |
use_spaces_gpu = False
|
|
|
12 |
def dummy_gpu_decorator(func):
|
13 |
return func
|
14 |
spaces = type('', (), {'GPU': dummy_gpu_decorator})()
|
15 |
+
|
16 |
# Define the GPTConfig class
|
17 |
class GPTConfig:
|
18 |
def __init__(self):
|
|
|
128 |
|
129 |
enc = tiktoken.get_encoding('gpt2')
|
130 |
|
131 |
+
|
132 |
# Update the generate_text function
|
133 |
+
@spaces.GPU
|
134 |
+
def generate_text(prompt, max_length=432, temperature=0.8, top_k=40):
|
|
|
135 |
model = load_model('gpt_model.pth')
|
136 |
device = next(model.parameters()).device
|
137 |
input_ids = torch.tensor(enc.encode(prompt)).unsqueeze(0).to(device)
|
|
|
150 |
input_ids = torch.cat([input_ids, next_token], dim=-1)
|
151 |
generated.append(next_token.item())
|
152 |
|
|
|
|
|
|
|
153 |
if next_token.item() == enc.encode('\n')[0] and len(generated) > 100:
|
154 |
break
|
|
|
|
|
155 |
|
156 |
+
return enc.decode(generated)
|
|
|
157 |
|
158 |
# Add the gradio_generate function
|
159 |
+
@spaces.GPU
|
160 |
+
def gradio_generate(prompt, max_length, temperature, top_k):
|
161 |
+
return generate_text(prompt, max_length, temperature, top_k)
|
|
|
|
|
|
|
162 |
|
163 |
|
164 |
# # Your existing imports and model code here...
|