File size: 2,426 Bytes
6a87471
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
---

license: mit
tags:
- pytorch
- tiny-transformer
- retrieval
---


# Tiny Transformer for Retrieval

## Overview

A research-oriented **Tiny Transformer** prototype targeting **Retrieval**. The included **giant** setup documents defaults and file formats without presenting unverified performance numbers.

## Repository status

- The Python file contains the model and runnable example or training entry point.
- `config.json` records the generated architecture settings.
- `training_args.json` records the default experiment recipe.
- `model.safetensors` is a valid initialization checkpoint for smoke tests; it is **not** presented as a trained benchmark checkpoint.
- No benchmark score is claimed in this repository.

## Architecture

| Item | Value |
|---|---|
| Architecture | Tiny Transformer |
| Scale | giant |
| Attention | sparse |
| Fusion | tensor fusion |
| Activation | relu |
| Normalization | layernorm |

## Default experiment recipe

The included configuration uses **lamb** with a **exponential** schedule. These are starting values in the script, not evidence of a completed run. For a meaningful evaluation, train all baselines with the same data exposure, tuning budget, and random seeds.

## Quick check

```bash

python model.py --help

```

Inspect the script's `__main__` block for its generated smoke-test example. Because this is a custom implementation, generic automatic loading APIs require an explicit adapter before use.

## Evaluation guidance

A useful first evaluation would use **Flickr30k**, report the task metric across at least three seeds, and include a matched-capacity baseline. Keep training logs and environment versions with any published result.

## Limitations

The initialization checkpoint has not been trained or audited for robustness, fairness, or domain transfer. The implementation should be treated as an experimental starting point. Results from a future trained checkpoint must be documented separately from the defaults shipped here.

## Files

- `model.py` — primary artifact
- `README.md` — this documentation
- `config.json` — architecture configuration
- `training_args.json` — default experiment settings
- `model.safetensors` — initialization checkpoint

## License

Released under **mit**. Review the source-data terms separately when this repository is used with external datasets.