Update app.py

app.py

@@ -69,15 +69,27 @@ def gguf_transformers(model_name, sl, tl, input_text):
 
 def gguf_ctransformers(model_name, sl, tl, input_text):
     from ctransformers import AutoModelForCausalLM
+    from huggingface_hub import hf_hub_download
+
+    # 1. Fetch the exact .gguf into your local cache
+    gguf_path = hf_hub_download(
+        repo_id="TheBloke/TowerInstruct-7B-v0.1-GGUF",
+        filename="TowerInstruct-7B-v0.1.Q4_K_M.gguf"
+    )
     
-    repo_id   = "TheBloke/TowerInstruct-7B-v0.1-GGUF"
-    filename  = "TowerInstruct-7B-v0.1.Q4_K_M.gguf"
+    # 2. Load once at module scope (so you don’t reload on every call)
     model = AutoModelForCausalLM.from_pretrained(
-        "https://huggingface.co/TheBloke/TowerInstruct-7B-v0.1-GGUF/resolve/main/towerinstruct-7b-v0.1.Q4_K_M.gguf",
+        gguf_path,
         model_type="llama"
     )
     prompt = f"Translate the following text from {sl} into {tl}.\n{sl}: {input_text}\n{tl}:"
-    return model.generate(prompt)[0]
+    out = model.generate(
+        prompt,
+        max_new_tokens=256,
+        temperature=0.2,
+        stop=[f"{tl}:"]
+    )
+    return out[0]
 
 def gguf(model_name, sl, tl, input_text):
     # 1. Pull in tokenizer (handles special tokens)
@@ -213,7 +225,7 @@ def translate_text(input_text: str, sselected_language: str, tselected_language:
         return translated_text, message_text
 
     if 'gguf' in model_name.lower():
-        translated_text = gguf_transformers(model_name, sselected_language, tselected_language, input_text)
+        translated_text = gguf_ctransformers(model_name, sselected_language, tselected_language, input_text)
         return translated_text, message_text
         
     if 'teuken' in model_name.lower():