GPU demo pending

Berkelium Design Planner

Demo of berkelium-qwen3-32b-lora, a LoRA adapter on Qwen3-32B for the Berkelium computational-engineering platform. The model only proposes; deterministic code computes and validates.

What the live demo does

Status

The Gradio + ZeroGPU app is written and included in this repo (app.py). It has not been run yet: a 32B model needs GPU hardware, and the account does not currently have access to ZeroGPU. Once GPU hardware is attached, set README.gradio.md as README.md to switch this Space to the live demo.

What it will look like

Reference only — the input and target below come from the test split of the dataset. This is the training target, not a live model output.

Request

Make two meshing spur gears at 90 mm centre distance, module 1 mm, for a gear ratio of 3.5.

DesignProposal (dataset target)
{
  "intent": {
    "assumptions": [],
    "extracted": [],
    "open_questions": [],
    "summary": "spur gear pair"
  },
  "kind": "design_proposal",
  "schema_version": "0.1.0",
  "specification": {
    "constraints": [],
    "parameters": [],
    "requirements": [
      {
        "applies_to": "pair",
        "comparator": "approx",
        "id": "r_ratio",
        "quantity": "ratio",
        "target": {
          "unit": "1",
          "value": 3.5
        },
        "tolerance": {
          "unit": "1",
          "value": 0.07
        }
      },
      {
        "applies_to": "pair",
        "comparator": "==",
        "id": "r_center_distance",
        "quantity": "center_distance",
        "target": {
          "unit": "mm",
          "value": 90.0
        },
        "tolerance": {
          "unit": "mm",
          "value": 0.01
        }
      },
      {
        "applies_to": "pair",
        "comparator": "==",
        "id": "r_module",
        "quantity": "module",
        "target": {
          "unit": "mm",
          "value": 1.0
        }
      }
    ]
  },
  "structure": {
    "components": [
      {
        "cem": "gear.spur_pair@0.1",
        "id": "pair",
        "kind": "cem",
        "requirements": {
          "center_distance": {
            "unit": "mm",
            "value": 90.0
          },
          "module": {
            "unit": "mm",
            "value": 1.0
          },
          "ratio": 3.5
        }
      }
    ],
    "relations": []
  }
}

Engineering validation is not_evaluated in this demo; it only checks that outputs are well-formed proposals.