Prepare Space for aim4composites T4 deployment

DEPLOYMENT.md

@@ -0,0 +1,51 @@
+# Hugging Face migration notes
+
+This repository was pulled from:
+
+- source Space: `https://huggingface.co/spaces/ChemFM/MaterialInverseDesignDemo`
+
+Target destination:
+
+- destination org: `https://huggingface.co/aim4composites`
+
+## What changed for `Nvidia T4 small`
+
+- README front matter now recommends `suggested_hardware: t4-small`
+- `spaces` was removed from `requirements.txt` because this deployment is intended for a dedicated GPU Space rather than ZeroGPU
+- the app text no longer describes the Space as ZeroGPU-only
+- runtime knobs were added so the thermoforming step can be tuned without code edits:
+  - `MG_DEFAULT_N_GENERATE`
+  - `MG_MAX_GENERATE`
+  - `MG_THERMO_RESTARTS`
+  - `MG_THERMO_EPOCHS`
+
+## Recommended Hugging Face Space settings
+
+1. Create a new Gradio Space under `aim4composites`.
+2. Push this repository to that new Space.
+3. In the Space hardware settings, choose `Nvidia T4 small`.
+4. Leave storage at the default unless larger artifacts are added later.
+5. After credits are added, restart the Space and verify logs show CUDA is available.
+
+## Suggested first validation pass
+
+1. Open the Space and load the default test condition.
+2. Run material inverse design with the default `3` candidates.
+3. Select the top-ranked design.
+4. Run thermoforming once with a small target example.
+5. Confirm:
+   - the app starts successfully
+   - candidate generation completes
+   - thermoforming completes
+   - no out-of-memory or timeout errors appear in the logs
+
+## Optional runtime tuning if the Space feels slow
+
+Set these Space variables:
+
+- `MG_DEFAULT_N_GENERATE=2`
+- `MG_MAX_GENERATE=4`
+- `MG_THERMO_RESTARTS=6`
+- `MG_THERMO_EPOCHS=300`
+
+If runtime remains comfortably fast, increase these gradually.

EMAIL_TO_DR_LI.md

@@ -0,0 +1,28 @@
+Subject: Request to enable Hugging Face credits for `aim4composites` Space deployment
+
+Hi Dr. Li,
+
+I migrated the `ChemFM/MaterialInverseDesignDemo` Hugging Face Space code so it is ready to be deployed under the `aim4composites` organization and configured for `Nvidia T4 small`.
+
+The target hardware we plan to use is:
+
+- `Nvidia T4 small`
+- `4 vCPU`
+- `15 GB RAM`
+- `16 GB VRAM`
+- currently listed by Hugging Face at `$0.40/hour`
+
+At the moment, I cannot fully validate the hosted deployment because the `aim4composites` organization does not yet have credits available for paid GPU hardware.
+
+Could you please purchase or enable enough Hugging Face credits for the organization so we can:
+
+1. assign `Nvidia T4 small` to the new Space,
+2. launch the app under the target hardware,
+3. run one end-to-end validation pass, and
+4. confirm the final deployment is stable.
+
+Once the credits are available, I can test the Space immediately and report back on startup, inference, and thermoforming performance.
+
+Best,
+
+[Your Name]

README.md

@@ -1,13 +1,17 @@
 ---
-title: Quarter V5 Inverse Design ZeroGPU
+title: Quarter V5 Inverse Design
 sdk: gradio
+python_version: "3.10"
 app_file: app.py
+suggested_hardware: t4-small
+startup_duration_timeout: 1h
 pinned: false
 ---
 
-# Quarter V5 Inverse Design ZeroGPU
+# Quarter V5 Inverse Design
 
-Local Gradio-first scaffold for a private Hugging Face Space under `ChemFM`.
+Local Gradio-first scaffold for migrating the Space from `ChemFM/MaterialInverseDesignDemo`
+to `aim4composites` on Hugging Face GPU hardware.
 
 Key points:
 
@@ -16,17 +20,22 @@ Key points:
 - keeps `vxy` and `vyx` as `2`-input requirements
 - generates `3` candidates and selects the closest one
 - runs thermoforming from the selected design
-- is structured for Hugging Face ZeroGPU by decorating the generation step with `@spaces.GPU`
+- is configured to run on a dedicated Hugging Face `Nvidia T4 small` Space
 
 ## Local run
 
 ```bash
-cd /project/luofeng/feiyang/MaterialGeneration/hf_space_generation_quarter_zerogpu_gradio
-/home/feiyang/miniconda3/envs/ChemFMNew/bin/python app.py
+python app.py
 ```
 
-## Notes for HF Space deployment
+## Notes for HF Space deployment on `t4-small`
 
-- ZeroGPU is Gradio-only and requires the `spaces` package plus a `@spaces.GPU` wrapper on GPU work.
 - This folder now bundles the quarter-v5 checkpoint, the minimal metadata/test files needed for inference, and the quarter simulation assets under `assets/`.
 - Local absolute paths are optional overrides through `MG_CHECKPOINT_DIR`, `MG_DATA_DIR`, `MG_CURVE_DIR`, and `MG_QUARTER_DATA_GEN_DIR`.
+- The Space recommends `t4-small` via `suggested_hardware`, but Hugging Face will not assign paid hardware automatically. The hardware still needs to be selected in the Space settings after credits are available.
+- Thermoforming latency is mostly CPU-bound. Dedicated tuning knobs are exposed through:
+  - `MG_DEFAULT_N_GENERATE`
+  - `MG_MAX_GENERATE`
+  - `MG_THERMO_RESTARTS`
+  - `MG_THERMO_EPOCHS`
+- Default settings are chosen to stay practical on `Nvidia T4 small` (`4 vCPU`, `15 GB RAM`, `16 GB VRAM`) while keeping the main inverse-design path on GPU.

app.py

@@ -60,9 +60,12 @@ CURVE_DIR = DEFAULT_CURVE_DIR
 NORMALIZATION_METHOD = "zscore"
 ANGLE_RESOLUTION = 1.0
 DEFAULT_N_GENERATE = 3
-MAX_N_GENERATE = 10
+MAX_N_GENERATE = int(os.environ.get("MG_MAX_GENERATE", "6"))
+DEFAULT_N_GENERATE = min(int(os.environ.get("MG_DEFAULT_N_GENERATE", str(DEFAULT_N_GENERATE))), MAX_N_GENERATE)
 GPU_DURATION_SEC = int(os.environ.get("ZEROGPU_DURATION", "90"))
 DEFAULT_TEST_INDEX = int(os.environ.get("MG_DEFAULT_TEST_INDEX", "512"))
+THERMO_N_RESTARTS = int(os.environ.get("MG_THERMO_RESTARTS", "6"))
+THERMO_EPOCHS = int(os.environ.get("MG_THERMO_EPOCHS", "300"))
 MATERIAL_TYPE_NAMES = {0: "CPP", 1: "CHDPE", 2: "GPP", 3: "GHDPE"}
 
 
@@ -588,8 +591,8 @@ def ui_thermoforming(selected_label: str, payload: str, angle_a: float, angle_b:
         ply_number=n_layers,
         fiber_vf=float(selected["vf"]),
         y_target=[float(angle_a), float(angle_b), float(angle_c), float(max_stress)],
-        n_restarts=10,
-        epochs=500,
+        n_restarts=THERMO_N_RESTARTS,
+        epochs=THERMO_EPOCHS,
     )
     warning_md = ""
     if domain_issues:
@@ -627,7 +630,7 @@ def build_app():
     with gr.Blocks(css=css) as demo:
         gr.Markdown("# Quarter Material Inverse Design")
         gr.Markdown(
-            "ZeroGPU-targeted Gradio app for the quarter-v5 model. "
+            "Dedicated-GPU Gradio app for the quarter-v5 model. "
             "Requirement curves are defined first, then batched candidates are generated, "
             "the closest design is selected, and thermoforming is run from that selected design."
         )

requirements.txt

@@ -5,6 +5,5 @@ numpy
 pandas
 openpyxl
 pyyaml
-spaces
 torch==2.8.0
 tqdm