Create app.py

app.py

@@ -0,0 +1,20 @@
+import os, subprocess, gradio as gr
+
+MODEL = "dolphin-2.9.3-llama3-8b.Q4_K_M.gguf"
+URL = "https://huggingface.co/TheBloke/dolphin-2.9.3-llama3-8b-GGUF/resolve/main/" + MODEL
+
+if not os.path.exists(MODEL):
+    subprocess.run(["wget", URL])
+
+if not os.path.exists("llama.cpp"):
+    subprocess.run(["git", "clone", "https://github.com/ggerganov/llama.cpp"])
+    subprocess.run(["make", "-C", "llama.cpp"])
+
+def chat(prompt):
+    p = subprocess.Popen(
+        ["./llama.cpp/main", "-m", MODEL, "-p", prompt, "-n", "200"],
+        stdout=subprocess.PIPE
+    )
+    return p.stdout.read().decode()
+
+gr.Interface(chat, gr.Textbox(label="Ask Dolphin"), gr.Textbox()).launch()