switch back to ollama

app.py

@@ -1,6 +1,11 @@
-from transformers import pipeline
+import subprocess
+#from transformers import pipeline
+import ollama
 import gradio
 
+
+subprocess.run("curl -fsSL https://ollama.com/install.sh | sh")
+
 history = []
 
 
@@ -13,8 +18,27 @@ def get_history_messages():
 
 
 def predict(prompt):
+    response = ollama.chat(
+        model="tinydolphin",
+        messages=[
+            *get_history_messages(),
+            {"role": "user", "content": prompt}
+        ],
+        stream=True
+    )
+    history.append((prompt, ""))
+    message = ""
+    for chunk in response:
+        message += chunk["message"]["content"]
+        history[-1] = (prompt, message)
+        yield "", history
+
+
+def predict_t(prompt):
     print("Predict:", prompt)
+    print("Loading model")
     pipe = pipeline("conversational", model="cognitivecomputations/TinyDolphin-2.8-1.1b")
+    print("Running pipeline")
     response = pipe(
         [
             *get_history_messages(),

requirements.txt

@@ -1,4 +1,3 @@
-torch
-transformers
+ollama
 gradio
 sentencepiece