Report ID: AIM-SEC-2024-13.7b
Subject: Mila AI -v1.3.7b- -aDDont-
Classification: Unstable / Emergent Behavior Observed
The keyword Mila AI -v1.3.7b- -aDDont- appears to be an unreleased or very niche model tag. While it evokes curiosity—hinting at adaptive mechanisms, responsible AI constraints, and a compact 1.3B parameter size—no verifiable implementation has been documented in public repositories.
For developers and researchers, this serves as a reminder to always include model cards, licenses, and example code when sharing novel AI artifacts. For enthusiasts, it’s an invitation to search custom Hugging Face spaces or contact Mila-affiliated researchers directly. Mila AI -v1.3.7b- -aDDont-
If you have access to this model or are its creator, please share a link in the discussion section below so this article can be updated with real benchmarks and usage examples.
Without official logs, we can estimate performance based on models of similar size: Report ID: AIM-SEC-2024-13
| Benchmark | Expected Score (1.3B) | Mila AI -v1.3.7b- -aDDont- (speculative) | |-----------|----------------------|-------------------------------------------| | HellaSwag (0-shot) | ~45% | ~48% (if well-tuned) | | MMLU (5-shot) | ~25% | ~27% | | HumanEval (pass@1) | ~4% | ~5.5% | | French GLUE (FLeX) | N/A | Could excel (bilingual) |
The -aDDont- might degrade or improve certain tasks depending on whether “don’t” refers to task-specific forgetting. Without official logs, we can estimate performance based
Given the parameter count (~1.37 billion), the model likely fits a decoder-only transformer similar to GPT-Neo, LLaMA‑small, or Phi‑1.5. Possible architecture choices:
| Component | Candidate Setting | |---------------------|---------------------------------------------| | Layers | 24–28 | | Hidden size | 2048–2560 | | Attention heads | 16–20 | | Context length | 2048 or 4096 tokens | | Activation function | SwiGLU / GELU | | Positional encoding | RoPE or ALiBi | | Training tokens | 300B – 1T (if scaled for 1.3B) |
The -aDDont- suffix might imply: