Transformers
PyTorch
English
language-model
graph-attention
adaptive-depth
temporal-decay
efficient-llm
Eval Results (legacy)
Instructions to use vigneshwar234/TemporalMesh-Transformer with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use vigneshwar234/TemporalMesh-Transformer with Transformers:
# Load model directly from transformers import AutoModel model = AutoModel.from_pretrained("vigneshwar234/TemporalMesh-Transformer", device_map="auto") - Notebooks
- Google Colab
- Kaggle
Add source: tmt/experiments/04_compare.ipynb
Browse files
tmt/experiments/04_compare.ipynb
ADDED
|
@@ -0,0 +1,61 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"cells": [
|
| 3 |
+
{
|
| 4 |
+
"cell_type": "markdown",
|
| 5 |
+
"metadata": {},
|
| 6 |
+
"source": ["# Experiment 04 — Perplexity Comparison Table\n", "Collects results from all ablations and renders a comparison table."]
|
| 7 |
+
},
|
| 8 |
+
{
|
| 9 |
+
"cell_type": "code",
|
| 10 |
+
"execution_count": null,
|
| 11 |
+
"metadata": {},
|
| 12 |
+
"outputs": [],
|
| 13 |
+
"source": [
|
| 14 |
+
"import pandas as pd\n",
|
| 15 |
+
"import matplotlib.pyplot as plt\n",
|
| 16 |
+
"\n",
|
| 17 |
+
"# Fill these in after running experiments 01-03\n",
|
| 18 |
+
"results = {\n",
|
| 19 |
+
" 'Model': ['Vanilla Transformer', 'Mesh Attention Only', 'Full TMT'],\n",
|
| 20 |
+
" 'Perplexity': [None, None, None], # replace with actual values\n",
|
| 21 |
+
" 'Avg Compute / Token': ['100%', '~60%', '~50%'],\n",
|
| 22 |
+
" 'Innovations': [\n",
|
| 23 |
+
" 'none',\n",
|
| 24 |
+
" 'mesh attention',\n",
|
| 25 |
+
" 'mesh + temporal decay + adaptive depth + dual-stream FFN + memory anchors'\n",
|
| 26 |
+
" ]\n",
|
| 27 |
+
"}\n",
|
| 28 |
+
"\n",
|
| 29 |
+
"df = pd.DataFrame(results)\n",
|
| 30 |
+
"print(df.to_markdown(index=False))"
|
| 31 |
+
]
|
| 32 |
+
},
|
| 33 |
+
{
|
| 34 |
+
"cell_type": "code",
|
| 35 |
+
"execution_count": null,
|
| 36 |
+
"metadata": {},
|
| 37 |
+
"outputs": [],
|
| 38 |
+
"source": [
|
| 39 |
+
"fig, ax = plt.subplots(figsize=(8, 4))\n",
|
| 40 |
+
"valid = [(m, p) for m, p in zip(results['Model'], results['Perplexity']) if p is not None]\n",
|
| 41 |
+
"if valid:\n",
|
| 42 |
+
" models, ppls = zip(*valid)\n",
|
| 43 |
+
" bars = ax.bar(models, ppls, color=['#4C72B0', '#DD8452', '#55A868'])\n",
|
| 44 |
+
" ax.bar_label(bars, fmt='%.1f')\n",
|
| 45 |
+
" ax.set_ylabel('Perplexity (lower is better)')\n",
|
| 46 |
+
" ax.set_title('TemporalMesh Transformer Ablation Study')\n",
|
| 47 |
+
" plt.tight_layout()\n",
|
| 48 |
+
" plt.savefig('ablation_perplexity.png', dpi=150)\n",
|
| 49 |
+
" plt.show()\n",
|
| 50 |
+
"else:\n",
|
| 51 |
+
" print('Run experiments 01-03 first, then fill in the Perplexity values above.')"
|
| 52 |
+
]
|
| 53 |
+
}
|
| 54 |
+
],
|
| 55 |
+
"metadata": {
|
| 56 |
+
"kernelspec": {"display_name": "Python 3", "language": "python", "name": "python3"},
|
| 57 |
+
"language_info": {"name": "python", "version": "3.10.0"}
|
| 58 |
+
},
|
| 59 |
+
"nbformat": 4,
|
| 60 |
+
"nbformat_minor": 4
|
| 61 |
+
}
|