---
license: mit
---
# 🚀 eblan-1.5-flash-lite

The fastest lightweight model fine-tuned on pure entropy.

### Technical Specifications:
- **Architecture:** EblanForCasualLM
- **Total Parameters:** 6,432,897 (~6.4M)
- **Trainable Parameters:** 1 (specifically `stupidity_param`)
- **Vocabulary Size:** ~50,257 (GPT-2 standard Byte-BPE)
- **Embedding Dimensions:** 64
- **Hidden Layers:** 1
- **Attention Heads:** 1

### Benchmark Results:
| Benchmark | Score |
| :--- | :---: |
| **AIME 2025** | `0.0` |
| **HLE** | `0.0` |
| **Terminal bench 2.1** | `0.0` |
| **SWE Bench verified** | `0.0` |

### Loss curve

![loss_curve](https://cdn-uploads.huggingface.co/production/uploads/6a1d885bbd8b36b69d8a8727/GtCPVtcer11_ODOEENVr2.png)