Update app.py

app.py

@@ -1,125 +1,34 @@
-import os
-import sys
-import ast
-
 import gradio as gr
-import torch
-import streamlit as st
-# No "spaces" or "transformers_gradio" imports here, since you said you want to use *your model* (myr1),
-# not external Spaces demos.
-
-from transformers import AutoTokenizer, AutoModelForCausalLM
-
-# ------------------------------------------------------------------------------
-# 1) OPTIONAL: Environment Variable Code (MY_SCRIPT_CONTENT)
-#    If you don't need this dynamic script execution, remove the entire block.
-# ------------------------------------------------------------------------------
-script_repr = os.getenv("MY_SCRIPT_CONTENT")
-
-if script_repr:
-    # Attempt to parse & exec the script from environment variable
-    try:
-        script_content = ast.literal_eval(script_repr)
-        exec(script_content)
-    except (ValueError, SyntaxError) as e:
-        # Using Streamlit to display an error message in case this is run within a Streamlit environment
-        st.error(f"Error evaluating script from environment variable: {e}")
-else:
-    print("No extra script content found in 'MY_SCRIPT_CONTENT'.")
-
-# ------------------------------------------------------------------------------
-# 2) Model References for "myr1" from Hugging Face
-#    Make sure your HF repo is "wuhp/myr1" and your actual model files are in subfolder "myr1"
-# ------------------------------------------------------------------------------
-MODEL_REPO = "wuhp/myr1"    # The HF repository name
-SUBFOLDER = "myr1"          # The folder inside the repo containing config.json etc.
-
-# ------------------------------------------------------------------------------
-# 3) Load Tokenizer & Model
-#    trust_remote_code=True to allow custom config/modeling if you have them in the repo.
-# ------------------------------------------------------------------------------
-print("Loading tokenizer...")
-tokenizer = AutoTokenizer.from_pretrained(
-    MODEL_REPO,
-    subfolder=SUBFOLDER,
-    trust_remote_code=True
+import spaces
+import transformers_gradio
+
+# Load *only* your model's interface. 
+# 
+# The original snippet loaded three models:
+#   demo = gr.load(name="deepseek-ai/DeepSeek-R1-Distill-Qwen-32B", src=transformers_gradio.registry)
+#   demo = gr.load(name="deepseek-ai/DeepSeek-R1", src=transformers_gradio.registry)
+#   demo = gr.load(name="deepseek-ai/DeepSeek-R1-Zero", src=transformers_gradio.registry)
+#
+# But we want the same UI, using *your* model from Hugging Face. So we do a single gr.load(...) call:
+# 
+# IMPORTANT:
+#   1) "name" should be the exact repository you want to load. 
+#   2) If your UI code was stored as a "Space" with a 'app.py' or 'api' in your "wuhp/myr1" repo, 
+#      this approach should pull that same Gradio interface. 
+#   3) If "transformers_gradio.registry" is correct for your space, keep it. 
+#      Otherwise, you might need "src='spaces'" or a different source, depending on how your space is set up.
+
+demo = gr.load(
+    name="wuhp/myr1", 
+    src="transformers_gradio.registry"
 )
 
-print("Loading model...")
-model = AutoModelForCausalLM.from_pretrained(
-    MODEL_REPO,
-    subfolder=SUBFOLDER,
-    trust_remote_code=True,
-    device_map="auto",         # auto-shard across GPU(s) if needed, else CPU fallback
-    torch_dtype=torch.float16,  # or torch.float32, torch.bfloat16, etc.
-    low_cpu_mem_usage=True
-)
-model.eval()
+# If you want GPU usage (like the original snippet):
+demo.fn = spaces.GPU()(demo.fn)
 
-print("Model loaded successfully.")
-
-# ------------------------------------------------------------------------------
-# 4) Define Generation Function for Gradio
-# ------------------------------------------------------------------------------
-def generate_text(prompt, max_new_tokens=64, temperature=0.7, top_p=0.9):
-    """
-    Generate text using the myr1 model from Hugging Face.
-    """
-    print("=== Starting generation ===")
-    inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
-
-    try:
-        output_ids = model.generate(
-            **inputs,
-            max_new_tokens=max_new_tokens,  # limit how many tokens beyond the prompt
-            temperature=temperature,
-            top_p=top_p,
-            do_sample=True,
-            pad_token_id=tokenizer.eos_token_id
-        )
-        print("=== Generation complete ===")
-    except Exception as e:
-        print(f"Error during generation: {e}")
-        return str(e)
-
-    return tokenizer.decode(output_ids[0], skip_special_tokens=True)
-
-
-# ------------------------------------------------------------------------------
-# 5) Build a Gradio UI
-# ------------------------------------------------------------------------------
-demo = gr.Interface(
-    fn=generate_text,
-    inputs=[
-        gr.Textbox(
-            lines=4,
-            label="Prompt",
-            placeholder="Ask a question or start a story..."
-        ),
-        gr.Slider(
-            minimum=8, maximum=512, step=1, value=64,
-            label="Max New Tokens"
-        ),
-        gr.Slider(
-            minimum=0.0, maximum=1.5, step=0.1, value=0.7,
-            label="Temperature"
-        ),
-        gr.Slider(
-            minimum=0.0, maximum=1.0, step=0.05, value=0.9,
-            label="Top-p (nucleus sampling)"
-        ),
-    ],
-    outputs="text",
-    title="DeepSeek myr1 Demo",
-    description=(
-        "Generates text with the 'myr1' model from the Hugging Face Hub. "
-        "Enter a prompt and adjust generation settings."
-    )
-)
+# Remove API names (like the original snippet):
+for fn in demo.fns.values():
+    fn.api_name = False
 
-# ------------------------------------------------------------------------------
-# 6) Launch the App
-# ------------------------------------------------------------------------------
 if __name__ == "__main__":
-    print("Launching Gradio demo...")
     demo.launch()