Initial CADFusion Space with Gradio

README.md

@@ -1,14 +1,6 @@
----
-title: CADFusionInterfaceProvided
-emoji: 📚
-colorFrom: green
-colorTo: gray
-sdk: gradio
-sdk_version: 5.44.0
-app_file: app.py
-pinned: false
-license: other
-short_description: Inference for Cad Fusion Model
----
+# CADFusion Hugging Face Space
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+This Space runs [Microsoft’s CADFusion](https://github.com/microsoft/CADFusion) on Hugging Face.  
+It loads the pretrained checkpoint `microsoft/CADFusion` and wraps it with a Gradio UI.  
+
+⚠️ **Note:** The full CADFusion model is large and GPU is required. A100 is recommended.  

app.py

@@ -1,61 +1,34 @@
-# app.py
 import gradio as gr
 import torch
-from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig
+from cadfusion.models import CADFusionModel
 
-MODEL_ID = "microsoft/CADFusion"
+# Load model (from HF weights + GitHub code)
+print("Loading CADFusion model...")
+device = "cuda" if torch.cuda.is_available() else "cpu"
 
-def load_model():
-    print("Loading tokenizer...")
-    tokenizer = AutoTokenizer.from_pretrained(MODEL_ID, use_fast=True)
-    print("Trying to load model in 4-bit (bitsandbytes)...")
-    try:
-        bnb_config = BitsAndBytesConfig(
-            load_in_4bit=True,
-            bnb_4bit_quant_type="nf4",
-            bnb_4bit_use_double_quant=True,
-            bnb_4bit_compute_dtype=torch.float16,
-        )
-        model = AutoModelForCausalLM.from_pretrained(
-            MODEL_ID,
-            quantization_config=bnb_config,
-            device_map="auto",
-            trust_remote_code=True,
-        )
-        print("Loaded in 4-bit")
-    except Exception as e:
-        print("4-bit load failed:", e)
-        print("Falling back to fp16 (may require larger GPU RAM)...")
-        model = AutoModelForCausalLM.from_pretrained(
-            MODEL_ID,
-            device_map="auto",
-            torch_dtype=torch.float16,
-            trust_remote_code=True,
-        )
-
-    model.eval()
-    return tokenizer, model
+model = CADFusionModel.from_pretrained("microsoft/CADFusion")
+model = model.to(device)
+model.eval()
 
-tokenizer, model = load_model()
-
-def generate(prompt, max_new_tokens=256):
-    if prompt is None or prompt.strip() == "":
-        return "Please provide a text description of the CAD model."
-    inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
+def generate(prompt):
+    """Run CADFusion inference on user prompt"""
     with torch.no_grad():
-        out = model.generate(**inputs, max_new_tokens=int(max_new_tokens), do_sample=False)
-    text = tokenizer.decode(out[0], skip_special_tokens=True)
-    return text
+        output = model.generate(
+            prompt,
+            max_new_tokens=256,
+            temperature=0.7,
+            top_p=0.9,
+        )
+    return output
 
+# Gradio UI
 with gr.Blocks() as demo:
-    gr.Markdown("# CADFusion demo (microsoft/CADFusion)\nEnter a design description and hit Generate.")
-    with gr.Row():
-        prompt = gr.Textbox(lines=5, placeholder="e.g. 'a coffee mug with cylindrical body and curved handle'")
-        tokens = gr.Slider(64, 1024, value=256, label="max_new_tokens")
-    out = gr.Textbox(lines=20)
+    gr.Markdown("## 🏗️ CADFusion Demo\nEnter a CAD prompt below:")
+    inp = gr.Textbox(label="Your CAD prompt")
+    out = gr.Textbox(label="Model Output")
+
     btn = gr.Button("Generate")
-    btn.click(fn=generate, inputs=[prompt, tokens], outputs=out)
+    btn.click(fn=generate, inputs=inp, outputs=out)
 
 if __name__ == "__main__":
     demo.launch()
-

requirements.txt

@@ -1,9 +1,5 @@
-gradio>=3.30
-transformers>=4.32
-bitsandbytes
 torch
+transformers
 accelerate
-safetensors
-sentencepiece
-huggingface-hub
-
+gradio
+git+https://github.com/microsoft/CADFusion.git