Spaces:
Build error
Build error
Update modules/inference.py
Browse files- modules/inference.py +3 -0
modules/inference.py
CHANGED
@@ -1,3 +1,5 @@
|
|
|
|
|
|
1 |
import os
|
2 |
import gradio as gr
|
3 |
import copy
|
@@ -10,6 +12,7 @@ import time
|
|
10 |
from modules.load_presets import load_presets_value
|
11 |
from modules.load_model import *
|
12 |
|
|
|
13 |
def generate_text(message, history, system_prompt, preset, temperature, max_tokens, top_p, top_k, repeat_penalty, model, n_ctx, n_gpu_layers, n_threads, verbose, f16_kv, logits_all, vocab_only, use_mmap, use_mlock, n_batch, last_n_tokens_size, low_vram, rope_freq_base, rope_freq_scale):
|
14 |
dir = os.getcwd()
|
15 |
global llm
|
|
|
1 |
+
import spaces
|
2 |
+
|
3 |
import os
|
4 |
import gradio as gr
|
5 |
import copy
|
|
|
12 |
from modules.load_presets import load_presets_value
|
13 |
from modules.load_model import *
|
14 |
|
15 |
+
@spaces.GPU
|
16 |
def generate_text(message, history, system_prompt, preset, temperature, max_tokens, top_p, top_k, repeat_penalty, model, n_ctx, n_gpu_layers, n_threads, verbose, f16_kv, logits_all, vocab_only, use_mmap, use_mlock, n_batch, last_n_tokens_size, low_vram, rope_freq_base, rope_freq_scale):
|
17 |
dir = os.getcwd()
|
18 |
global llm
|