| { | |
| "format": "torch-dimensions/architecture", | |
| "io": { | |
| "cells_per_step": 64, | |
| "input": [ | |
| "B", | |
| 8, | |
| 8, | |
| 96 | |
| ], | |
| "output": [ | |
| "B", | |
| 8, | |
| 8, | |
| 96 | |
| ] | |
| }, | |
| "lattice": { | |
| "axes": [ | |
| { | |
| "dynamic": false, | |
| "name": "row", | |
| "size": 8 | |
| }, | |
| { | |
| "dynamic": false, | |
| "name": "col", | |
| "size": 8 | |
| } | |
| ], | |
| "cells": { | |
| "dense": true, | |
| "present": 64, | |
| "present_rle": [ | |
| 0, | |
| 64 | |
| ], | |
| "total": 64 | |
| }, | |
| "n_axes": 2, | |
| "names": [ | |
| "row", | |
| "col" | |
| ], | |
| "rank": 2, | |
| "shape": [ | |
| 8, | |
| 8 | |
| ], | |
| "time": false | |
| }, | |
| "layers": [ | |
| { | |
| "axes": [ | |
| "row", | |
| "col" | |
| ], | |
| "axis": null, | |
| "axis_index": null, | |
| "joins_time": false, | |
| "kind": "flatten", | |
| "layer": 0, | |
| "mixer": "AttentionMixer", | |
| "n_params": 74496, | |
| "reverse": false, | |
| "tokens": 64 | |
| }, | |
| { | |
| "axes": [ | |
| "row", | |
| "col" | |
| ], | |
| "axis": null, | |
| "axis_index": null, | |
| "joins_time": false, | |
| "kind": "flatten", | |
| "layer": 1, | |
| "mixer": "AttentionMixer", | |
| "n_params": 74496, | |
| "reverse": false, | |
| "tokens": 64 | |
| }, | |
| { | |
| "axes": [ | |
| "row", | |
| "col" | |
| ], | |
| "axis": null, | |
| "axis_index": null, | |
| "joins_time": false, | |
| "kind": "flatten", | |
| "layer": 2, | |
| "mixer": "AttentionMixer", | |
| "n_params": 74496, | |
| "reverse": false, | |
| "tokens": 64 | |
| }, | |
| { | |
| "axes": [ | |
| "row", | |
| "col" | |
| ], | |
| "axis": null, | |
| "axis_index": null, | |
| "joins_time": false, | |
| "kind": "flatten", | |
| "layer": 3, | |
| "mixer": "AttentionMixer", | |
| "n_params": 74496, | |
| "reverse": false, | |
| "tokens": 64 | |
| }, | |
| { | |
| "axes": [ | |
| "row", | |
| "col" | |
| ], | |
| "axis": null, | |
| "axis_index": null, | |
| "joins_time": false, | |
| "kind": "flatten", | |
| "layer": 4, | |
| "mixer": "AttentionMixer", | |
| "n_params": 74496, | |
| "reverse": false, | |
| "tokens": 64 | |
| }, | |
| { | |
| "axes": [ | |
| "row", | |
| "col" | |
| ], | |
| "axis": null, | |
| "axis_index": null, | |
| "joins_time": false, | |
| "kind": "flatten", | |
| "layer": 5, | |
| "mixer": "AttentionMixer", | |
| "n_params": 74496, | |
| "reverse": false, | |
| "tokens": 64 | |
| } | |
| ], | |
| "model": { | |
| "d_input": 96, | |
| "d_model": 96, | |
| "kind": "ViT", | |
| "n_layers": 6, | |
| "n_params": 454368 | |
| }, | |
| "nd_method": { | |
| "family": "flatten", | |
| "name": "Flatten" | |
| }, | |
| "sweeps": { | |
| "contracted_axes": [], | |
| "coverage": null, | |
| "directions": {}, | |
| "joint_axes": [ | |
| "row", | |
| "col" | |
| ], | |
| "pinned_axes": [], | |
| "unswept_axes": [] | |
| }, | |
| "version": 2 | |
| } | |