Spaces:
Sleeping
Sleeping
Initial submission of Analog Town to Build Small Hackathon
Browse files- SUBMISSION_BLOG.md +0 -35
SUBMISSION_BLOG.md
CHANGED
|
@@ -40,38 +40,3 @@ To comply with the **Build Small Hackathon** constraints, we engineered the app
|
|
| 40 |
2. **Deterministic State Transitions**: Rather than standard free-form generation, we model minds as discrete Pydantic structures (see schemas.py). The model outputs valid JSON conforming to:
|
| 41 |
$$\text{Previous State} + \text{Broadcast Event} \rightarrow \text{Updated State} + \text{Monologue} + \text{Likely Actions}$$
|
| 42 |
3. **Local Dataset Exports**: Traces are collected, cleaned of PII, packed, and uploaded as structured datasets to the Hugging Face Hub using export_hub.py.
|
| 43 |
-
|
| 44 |
-
---
|
| 45 |
-
|
| 46 |
-
## 🎬 Demo Video Recording Script
|
| 47 |
-
|
| 48 |
-
*Use this script as a guide to record your 1-2 minute demo video for submission:*
|
| 49 |
-
|
| 50 |
-
1. **Introduction (0:00 - 0:15)**:
|
| 51 |
-
- Start with the app loaded on **Graybridge Town**.
|
| 52 |
-
- Hover over the characters on the map to show their names and roles.
|
| 53 |
-
- Explain the premise: *"This is Analog Town, a shortwave receiver simulator for fictional minds. We have loaded Graybridge Town, a quiet rail community."*
|
| 54 |
-
|
| 55 |
-
2. **The Dossier (0:15 - 0:30)**:
|
| 56 |
-
- Click on **Arthur Pendelton** (the bearded senior male avatar).
|
| 57 |
-
- Point out how the dossiers load his core values, wounds, fears, and hopes.
|
| 58 |
-
- Show that his radio frequency is **89.1 FM**.
|
| 59 |
-
|
| 60 |
-
3. **Broadcast Event (0:30 - 0:50)**:
|
| 61 |
-
- Enter a broadcast event:
|
| 62 |
-
- **Title**: *Luxury Hotel Conversion*
|
| 63 |
-
- **Content**: *The historic brick train depot will be demolished next month to build a high-rise luxury hotel.*
|
| 64 |
-
- Click **🔊 TRANSMIT TO TOWN**.
|
| 65 |
-
- Show the transmission log printing status lines as it updates each character's mind.
|
| 66 |
-
|
| 67 |
-
4. **Tuning the Dial (0:50 - 1:20)**:
|
| 68 |
-
- Slide the frequency slider to **89.1 FM**.
|
| 69 |
-
- Show Arthur's monologue loading: *"My grandfather helped build those tracks... now they want to turn it into a playground for tourists."*
|
| 70 |
-
- Point out the active green signal strength bars and how his emotional bars (Anger, Fear) spiked.
|
| 71 |
-
- Slide the dial back to **88.0 FM** to show the static returning.
|
| 72 |
-
- Tune to **92.3 FM** (Elena Vance) to hear a different perspective.
|
| 73 |
-
|
| 74 |
-
5. **Dataset Export (1:20 - 1:40)**:
|
| 75 |
-
- Open the **Export & Share** accordion.
|
| 76 |
-
- Explain that all structured simulation traces can be exported as a Hugging Face dataset.
|
| 77 |
-
- Conclude: *"Analog Town shows how we can use small language models (<32B) under structured schemas to simulate social weather patterns."*
|
|
|
|
| 40 |
2. **Deterministic State Transitions**: Rather than standard free-form generation, we model minds as discrete Pydantic structures (see schemas.py). The model outputs valid JSON conforming to:
|
| 41 |
$$\text{Previous State} + \text{Broadcast Event} \rightarrow \text{Updated State} + \text{Monologue} + \text{Likely Actions}$$
|
| 42 |
3. **Local Dataset Exports**: Traces are collected, cleaned of PII, packed, and uploaded as structured datasets to the Hugging Face Hub using export_hub.py.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|