Upload 2 files

dc.py

@@ -538,10 +538,6 @@ class GuiSD:
                 if save_generated_images:
                     info_images += f"<br>{download_links}"
 
-## BEGIN MOD
-                img = save_images(img, metadata)
-## END MOD
-
                 info_state = "COMPLETE"
 
             yield info_state, img, info_images
@@ -710,7 +706,7 @@ def infer(prompt, negative_prompt, seed, randomize_seed, width, height, guidance
 
     progress(0, desc="Loading model...")
     for m in sd_gen.load_new_model(model_name, vae, TASK_MODEL_LIST[0]):
-        progress(0.5, desc=m)
+        print(m)
     progress(1, desc="Model loaded.")
     progress(0, desc="Starting Inference...")
     images = None
@@ -726,7 +722,7 @@ def infer(prompt, negative_prompt, seed, randomize_seed, width, height, guidance
         True, None, None, "plus_face", "original", 0.7, None, None, "base", "style", 0.7, 0.0,
         load_lora_cpu, verbose_info, gpu_duration
     ):
-        progress(0.5, desc=info_state)
+        print(info_state)
         images = stream_images
     progress(1, desc="Inference completed.")
     output_image = images[0][0] if images else None

llmdolphin.py

@@ -1,5 +1,9 @@
 import spaces
 import gradio as gr
+from pathlib import Path
+import re
+import torch
+from huggingface_hub import hf_hub_download, HfApi
 from llama_cpp import Llama
 from llama_cpp_agent import LlamaCppAgent, MessagesFormatterType
 from llama_cpp_agent.providers import LlamaCppPythonProvider
@@ -7,7 +11,6 @@ from llama_cpp_agent.chat_history import BasicChatHistory
 from llama_cpp_agent.chat_history.messages import Roles
 from ja_to_danbooru.ja_to_danbooru import jatags_to_danbooru_tags
 import wrapt_timeout_decorator
-from pathlib import Path
 from llama_cpp_agent.messages_formatter import MessagesFormatter
 from formatter import mistral_v1_formatter, mistral_v2_formatter, mistral_v3_tekken_formatter
 
@@ -846,6 +849,7 @@ llm_languages = ["English", "Japanese", "Chinese", "Korean", "Spanish", "Portugu
 llm_models_tupled_list = []
 default_llm_model_filename = list(llm_models.keys())[0]
 override_llm_format = None
+device = "cuda" if torch.cuda.is_available() else "cpu"
 
 
 def to_list(s):
@@ -858,7 +862,6 @@ def list_uniq(l):
 
 @wrapt_timeout_decorator.timeout(dec_timeout=3.5)
 def to_list_ja(s):
-    import re
     s = re.sub(r'[、。]', ',', s)
     return [x.strip() for x in s.split(",") if not s == ""]
 
@@ -873,7 +876,6 @@ def is_japanese(s):
 
 
 def update_llm_model_tupled_list():
-    from pathlib import Path
     global llm_models_tupled_list
     llm_models_tupled_list = []
     for k, v in llm_models.items():
@@ -890,7 +892,6 @@ def update_llm_model_tupled_list():
 
 
 def download_llm_models():
-    from huggingface_hub import hf_hub_download
     global llm_models_tupled_list
     llm_models_tupled_list = []
     for k, v in llm_models.items():
@@ -904,7 +905,6 @@ def download_llm_models():
 
 
 def download_llm_model(filename):
-    from huggingface_hub import hf_hub_download
     if not filename in llm_models.keys(): return default_llm_model_filename
     try:
         hf_hub_download(repo_id = llm_models[filename][0], filename = filename, local_dir = llm_models_dir)
@@ -965,8 +965,6 @@ def get_dolphin_model_format(filename):
 
 
 def add_dolphin_models(query, format_name):
-    import re
-    from huggingface_hub import HfApi
     global llm_models
     api = HfApi()
     add_models = {}
@@ -991,7 +989,6 @@ def add_dolphin_models(query, format_name):
     except Exception as e:
         print(e)
         return gr.update(visible=True)
-    #print(add_models)
     llm_models = (llm_models | add_models).copy()
     update_llm_model_tupled_list()
     choices = get_dolphin_models()
@@ -1191,7 +1188,6 @@ Output should be enclosed in //GENBEGIN//:// and //://GENEND//. The text to be g
 
 
 def get_dolphin_sysprompt():
-    import re
     prompt = re.sub('<LANGUAGE>', dolphin_output_language, dolphin_system_prompt.get(dolphin_sysprompt_mode, ""))
     return prompt
 
@@ -1221,11 +1217,11 @@ def select_dolphin_language(lang: str):
 
 @wrapt_timeout_decorator.timeout(dec_timeout=5.0)
 def get_raw_prompt(msg: str):
-    import re
     m = re.findall(r'/GENBEGIN/(.+?)/GENEND/', msg, re.DOTALL)
     return re.sub(r'[*/:_"#\n]', ' ', ", ".join(m)).lower() if m else ""
 
 
+@torch.inference_mode()
 @spaces.GPU(duration=60)
 def dolphin_respond(
     message: str,
@@ -1239,7 +1235,6 @@ def dolphin_respond(
     repeat_penalty: float = 1.1,
     progress=gr.Progress(track_tqdm=True),
 ):
-    from pathlib import Path
     progress(0, desc="Processing...")
 
     if override_llm_format:
@@ -1320,6 +1315,7 @@ def dolphin_parse(
     return ", ".join(prompts), gr.update(interactive=True), gr.update(interactive=True)
 
 
+@torch.inference_mode()
 @spaces.GPU(duration=60)
 def dolphin_respond_auto(
     message: str,
@@ -1334,7 +1330,6 @@ def dolphin_respond_auto(
     progress=gr.Progress(track_tqdm=True),
 ):
     #if not is_japanese(message): return [(None, None)]
-    from pathlib import Path
     progress(0, desc="Processing...")
 
     if override_llm_format:
@@ -1391,7 +1386,7 @@ def dolphin_respond_auto(
     )
 
     progress(0.5, desc="Processing...")
-    
+
     outputs = ""
     for output in stream:
         outputs += output
@@ -1421,6 +1416,8 @@ def dolphin_parse_simple(
 import cv2
 cv2.setNumThreads(1)
 
+
+@torch.inference_mode()
 @spaces.GPU()
 def respond_playground(
     message,