Spaces:
Paused
Paused
Update app.py
Browse files
app.py
CHANGED
@@ -5,20 +5,30 @@ def gpu():
|
|
5 |
|
6 |
import os
|
7 |
import subprocess
|
|
|
8 |
import gradio as gr
|
9 |
|
10 |
MODEL_PATH = os.environ.get("MODEL_PATH", "models/BitNet-b1.58-2B-4T/ggml-model-i2_s.gguf")
|
11 |
|
12 |
def generate(prompt, max_tokens=128, temperature=0.7):
|
|
|
13 |
cmd = [
|
14 |
-
|
|
|
15 |
"-m", MODEL_PATH,
|
16 |
"-p", prompt,
|
17 |
"-n", str(max_tokens),
|
18 |
"-temp", str(temperature)
|
19 |
]
|
20 |
-
|
21 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
22 |
|
23 |
iface = gr.Interface(
|
24 |
fn=generate,
|
|
|
5 |
|
6 |
import os
|
7 |
import subprocess
|
8 |
+
import sys
|
9 |
import gradio as gr
|
10 |
|
11 |
MODEL_PATH = os.environ.get("MODEL_PATH", "models/BitNet-b1.58-2B-4T/ggml-model-i2_s.gguf")
|
12 |
|
13 |
def generate(prompt, max_tokens=128, temperature=0.7):
|
14 |
+
cwd = os.path.dirname(__file__)
|
15 |
cmd = [
|
16 |
+
sys.executable,
|
17 |
+
os.path.join(cwd, "run_inference.py"),
|
18 |
"-m", MODEL_PATH,
|
19 |
"-p", prompt,
|
20 |
"-n", str(max_tokens),
|
21 |
"-temp", str(temperature)
|
22 |
]
|
23 |
+
proc = subprocess.run(
|
24 |
+
cmd,
|
25 |
+
capture_output=True,
|
26 |
+
text=True,
|
27 |
+
cwd=cwd
|
28 |
+
)
|
29 |
+
if proc.returncode != 0:
|
30 |
+
return f"error: {proc.stderr.strip()}"
|
31 |
+
return proc.stdout.strip()
|
32 |
|
33 |
iface = gr.Interface(
|
34 |
fn=generate,
|