Update app.py

app.py

@@ -23,7 +23,7 @@ from io import StringIO
 
 last = 0
 CHROMA_DATA_PATH = "chroma_data/"
-EMBED_MODEL = "BAAI/bge-m3"
+EMBED_MODEL = "sentence-transformers/all-MiniLM-L6-v2" #"BAAI/bge-m3"
 # all-MiniLM-L6-v2
 CHUNK_SIZE = 800
 CHUNK_OVERLAP = 50
@@ -31,10 +31,12 @@ max_results = 3
 min_len = 40
 min_distance = 0.35
 max_distance = 0.6
-temperature = 0.55
-max_tokens=3072
+temperature = 0.6
+max_tokens=5100
 top_p=0.8
+top_k=1000
 frequency_penalty=0.0
+repetition_penalty=1.12
 presence_penalty=0.15
 jezik = "srpski"
 cs = "s0"
@@ -66,10 +68,12 @@ system_propmpt = system_sr
 Settings.llm = HuggingFaceInferenceAPI(model_name="mistralai/Mistral-Nemo-Instruct-2407",
                               device_map="auto",
                               system_prompt = system_propmpt,
-                              context_window=5000,
-                              max_new_tokens=2048,
+                              context_window=5100,
+                              max_new_tokens=3072,
 #                              stopping_ids=[50278, 50279, 50277, 1, 0],
-                              generate_kwargs={"temperature": 0.55, "top_p":0.92, "repetition_penalty":1.10, "do_sample": False},
+                              generate_kwargs={"temperature": temperature, "top_p":top_p, "repetition_penalty": repetition_penalty,
+                                               "presence_penalty": presence_penalty, "frequency_penalty": frequency_penalty,
+                                               "top_k": top_k, "do_sample": False},
 #                              tokenizer_kwargs={"max_length": 4096},
                               tokenizer_name="mistralai/Mistral-Nemo-Instruct-2407", 
                              )
@@ -164,13 +168,13 @@ with gr.Blocks() as iface:
         title="UChat",
         description="Postavite pitanje ili opišite problem koji imate",
         chatbot=gr.Chatbot(placeholder="Kako Vam mogu pomoći?", type="tuples", label="Agent podrške", height=400),
-        textbox=gr.Textbox(placeholder="Pitanje ili opis problema", container=True, scale=7),
+        textbox=gr.Textbox(placeholder="Pitanje ili opis problema", container=False, scale=7),
         theme="soft",
 #        examples=["Ne radi mi internet", "Koje usluge imam na raspologanju?", "Ne radi mi daljinski upravljač, šta da radim?"],
 #        cache_examples=True,
         retry_btn=None,
         undo_btn=None,
-        clear_btn="Briši sve",
+        clear_btn="Briši sve - razgovor ispočetka",
         additional_inputs = [gr.Dropdown(["slovenski", "hrvatski", "srpski", "makedonski", "Eksperimentalna opcija"], value="srpski", label="Jezik", info="N/A"),
                              gr.File()
                          ],