Upload app.py with huggingface_hub

app.py

@@ -6,12 +6,8 @@ import gradio as gr
 from transformers import AutoModelForCausalLM, AutoTokenizer
 from peft import PeftModel
 
-MODELS = {
-    "Qwen3-8B": {
-        "base": "Qwen/Qwen3-8B",
-        "adapter": "hikewa/dialectic-qwen3-8b-lora",
-    },
-}
+BASE_MODEL = "Qwen/Qwen3-8B"
+ADAPTER = "hikewa/dialectic-qwen3-8b-lora"
 
 SYSTEM_PROMPT = (
     "You reason carefully through problems by considering competing "
@@ -20,40 +16,32 @@ SYSTEM_PROMPT = (
     "integrate insights rather than picking sides or hedging."
 )
 
-loaded = {"name": None, "model": None, "tokenizer": None}
+loaded = {"model": None, "tokenizer": None}
 
 
-def load_model(model_name):
+def load_model():
     global loaded
-    if loaded["name"] == model_name:
-        return loaded["model"], loaded["tokenizer"]
-
-    # Free previous model
     if loaded["model"] is not None:
-        del loaded["model"]
-        loaded["model"] = None
-        torch.cuda.empty_cache()
+        return loaded["model"], loaded["tokenizer"]
 
-    cfg = MODELS[model_name]
     tokenizer = AutoTokenizer.from_pretrained(
-        cfg["adapter"], trust_remote_code=True
+        ADAPTER, trust_remote_code=True
     )
     base = AutoModelForCausalLM.from_pretrained(
-        cfg["base"], torch_dtype=torch.float16, trust_remote_code=True
+        BASE_MODEL, torch_dtype=torch.float16, trust_remote_code=True
     )
-    model = PeftModel.from_pretrained(base, cfg["adapter"])
+    model = PeftModel.from_pretrained(base, ADAPTER)
     model = model.to("cuda")
     model.eval()
 
-    loaded["name"] = model_name
     loaded["model"] = model
     loaded["tokenizer"] = tokenizer
     return model, tokenizer
 
 
 @spaces.GPU
-def respond(message, history, model_name):
-    model, tokenizer = load_model(model_name)
+def respond(message, history):
+    model, tokenizer = load_model()
 
     messages = [{"role": "system", "content": SYSTEM_PROMPT}]
     for msg in history:
@@ -87,24 +75,18 @@ def respond(message, history, model_name):
 
 demo = gr.ChatInterface(
     respond,
-    additional_inputs=[
-        gr.Dropdown(
-            choices=list(MODELS.keys()),
-            value="Qwen3-8B",
-            label="Model",
-        ),
-    ],
-    title="Dialectic Reasoning Models",
+    title="Dialectic Reasoning",
     description=(
-        "Fine-tuned on 510 dialectic reasoning traces. "
+        "Qwen3-8B fine-tuned on 510 dialectic reasoning traces. "
         "Ask a question involving competing perspectives."
     ),
     examples=[
-        ["Should AI systems be transparent about their reasoning, even when transparency reduces performance?"],
-        ["Is it better to optimize for individual freedom or collective wellbeing?"],
-        ["When does pragmatic compromise become unprincipled capitulation?"],
+        "Should AI systems be transparent about their reasoning, even when transparency reduces performance?",
+        "Is it better to optimize for individual freedom or collective wellbeing?",
+        "When does pragmatic compromise become unprincipled capitulation?",
     ],
+    cache_examples=False,
 )
 
 if __name__ == "__main__":
-    demo.launch()
+    demo.launch(ssr_mode=False)