Try v3 (Get rid of LoRA stuff)

app.py

@@ -34,7 +34,7 @@ def start_worker(model_path: str, bits=16):
     assert bits in [4, 8, 16], "It can be only loaded with 16-bit, 8-bit, and 4-bit."
     if bits != 16:
         model_name += f"-{bits}bit"
-    model_name += "-lora"
+    # model_name += "-lora"
     worker_command = [
         sys.executable,
         "-m",
@@ -47,8 +47,6 @@ def start_worker(model_path: str, bits=16):
         model_path,
         "--model-name",
         model_name,
-        "--model-base",
-        "liuhaotian/llava-1.5-7b",
         "--use-flash-attn",
     ]
     # if bits != 16:
@@ -74,8 +72,8 @@ if __name__ == "__main__":
 
     print(f"astro args: {gws.args}")
 
-    model_path = os.getenv("model", "universeTBD/AstroLLaVA_v2")
-    bits = int(os.getenv("bits", 4))
+    model_path = os.getenv("model", "universeTBD/AstroLLaVA_v3")
+    bits = int(os.getenv("bits", 16))
     concurrency_count = int(os.getenv("concurrency_count", 5))
 
     controller_proc = start_controller()