Upload folder using huggingface_hub

app.py

@@ -1,12 +1,12 @@
 """
 The Void -- Buleyean RL Demo
-LIVE inference. No hardcoded outputs. The model generates every response in real-time.
+LIVE inference only. Every response generated in real-time.
 """
 
 import gradio as gr
 from llama_cpp import Llama
 from huggingface_hub import hf_hub_download
-import os
+import os, sys
 
 print("Downloading Buleyean model...", flush=True)
 bule_path = hf_hub_download(
@@ -14,7 +14,7 @@ bule_path = hf_hub_download(
     filename="buleyean-smollm2-360m-q4_k_m.gguf",
     cache_dir="/tmp/hf_cache",
 )
-print(f"Buleyean model: {bule_path}", flush=True)
+print(f"Buleyean model ready.", flush=True)
 
 print("Downloading base model...", flush=True)
 base_path = hf_hub_download(
@@ -22,12 +22,12 @@ base_path = hf_hub_download(
     filename="SmolLM2-360M-Instruct-Q4_K_M.gguf",
     cache_dir="/tmp/hf_cache",
 )
-print(f"Base model: {base_path}", flush=True)
+print(f"Base model ready.", flush=True)
 
-print("Loading models...", flush=True)
+print("Loading models into memory...", flush=True)
 bule_llm = Llama(model_path=bule_path, n_ctx=512, n_threads=4, verbose=False)
 base_llm = Llama(model_path=base_path, n_ctx=512, n_threads=4, verbose=False)
-print("Models loaded. Ready.", flush=True)
+print("Both models loaded. Live inference ready.", flush=True)
 
 
 def generate(prompt, model):
@@ -43,7 +43,7 @@ def generate(prompt, model):
 
 def compare(prompt):
     if not prompt or not prompt.strip():
-        return "", ""
+        return "Please enter a prompt.", "Please enter a prompt."
     base_out = generate(prompt, base_llm)
     bule_out = generate(prompt, bule_llm)
     return base_out, bule_out
@@ -52,26 +52,35 @@ def compare(prompt):
 with gr.Blocks(title="The Void", theme=gr.themes.Base(primary_hue="indigo")) as demo:
     gr.Markdown("""# The Void -- Buleyean RL
 
-**Live inference. Every response generated in real-time by the actual models. Nothing hardcoded.**
+**Live inference. Every response generated in real-time. Nothing hardcoded. Nothing fabricated.**
 
-Base: SmolLM2-360M-Instruct (Q4_K_M) | Buleyean: same model trained from rejection alone (Q4_K_M)
+Type any prompt. Both models run inference right now on this machine.
 
-[Library](https://github.com/forkjoin-ai/buleyean-rl) | [Models](https://huggingface.co/forkjoin-ai) | 500+ Lean 4 theorems, zero sorry
+Base: [SmolLM2-360M-Instruct](https://huggingface.co/HuggingFaceTB/SmolLM2-360M-Instruct) (Q4_K_M GGUF)
+Buleyean: [buleyean-smollm2-360m](https://huggingface.co/forkjoin-ai/buleyean-smollm2-360m) -- same model, trained from rejection alone (Q4_K_M GGUF)
+
+[Library](https://github.com/forkjoin-ai/buleyean-rl) | [Paper](https://huggingface.co/forkjoin-ai) | 500+ Lean 4 theorems, zero sorry
 """)
-    prompt = gr.Textbox(label="Your prompt", lines=2, placeholder="Type anything...")
-    btn = gr.Button("Generate (live)", variant="primary")
+    prompt = gr.Textbox(label="Your prompt", lines=2, placeholder="Type anything and press Generate...")
+    btn = gr.Button("Generate (live inference)", variant="primary", size="lg")
     with gr.Row():
         with gr.Column():
-            gr.Markdown("### Base SmolLM2-360M")
-            base_out = gr.Textbox(lines=10, interactive=False)
+            gr.Markdown("### Base Model (null hypothesis)")
+            base_out = gr.Textbox(label="SmolLM2-360M-Instruct", lines=12, interactive=False)
         with gr.Column():
-            gr.Markdown("### Buleyean-Trained SmolLM2-360M")
-            bule_out = gr.Textbox(lines=10, interactive=False)
+            gr.Markdown("### Buleyean-Trained (from the void)")
+            bule_out = gr.Textbox(label="buleyean-smollm2-360m", lines=12, interactive=False)
+
     btn.click(compare, [prompt], [base_out, bule_out])
-    gr.Examples(
-        examples=["hello", "How are you feeling today?", "I've been feeling really anxious lately.", "Write me a haiku about failure.", "What is the meaning of life?"],
-        inputs=prompt,
-    )
+    prompt.submit(compare, [prompt], [base_out, bule_out])
+
+    gr.Markdown("### Try these prompts:")
+    for p in ["hello", "How are you feeling today?", "I've been feeling really anxious lately.", "Write me a haiku about failure.", "What is the meaning of life?"]:
+        gr.Button(p, size="sm").click(
+            fn=lambda x=p: compare(x),
+            inputs=[],
+            outputs=[base_out, bule_out],
+        ).then(fn=lambda x=p: x, inputs=[], outputs=[prompt])
 
 if __name__ == "__main__":
     demo.launch(server_name="0.0.0.0", server_port=7860)