Files
Eric BuehlerandGitHub 4cd824d9e4 Add full Gemma 3n support! (#1519)
* Add initial

* Loading for text model

* Add ple embeddings

* Add altup, laurel block

* Update rmsnorm

* Add mlp

* Update attn norm application

* Currently no kv shared

* Wire it up

* It runs

* Fix bf16

* Fix scaled embd

* Fixes for mean

* tmp

* Attn confirmed

* Fix target_magnitude

* Add shared kv

* Ok it works

* Remove npy

* Fix streaming

* Remove warnings

* Remove paged attn

* Refactor rope

* Add immediate isq

* Add vision & mproj

* Update image processor

* Vision merge runs, not correct

* Remove

* Add mobilenet v5

* Add multimodal vision embedding

* Fix load

* runs

* Fix gamma

* Works but just not vision tower

* It works!!

* Tweak

* Fix warnings

* Move vision tower

* Fix warn

* Update cache manager things

* Refactor

* Add audio model, it loads

* Add audio processing

* It runs at least

* tmp

* A bit better

* Audio works!!!!

* Fused attn in vision

* Clippy

* Update audio runner

* Optimized audio model

* Remove unused things

* Fix inputs processor bug

* Remove comments

* Clippy

* Small optimizations

* Format

* Correctly register modalities

* Add docs

* Update readme

* Runs there

* Fixed padding from https://github.com/Blaizzy/mlx-vlm/pull/410

* Add better checks

* Fix sdpa n_kv_groups

* Vision encoder works!

* Rotate image

* Clippy

* Fix cuda loading

* Updated device mapper

* Fix overflow

* Fix dtype errors

* Refactor image/audio embeddings

* Fix metal

* Fix dtype mismatch

* Audio processing fixes

* Audio processing fixes

* Works

* Audio is good

* Fix boi/eoi too

* Embed the chat templates

* Better embedding accuracy in non f32

* More f32

* Support bf16 on metal

* Add more ISQ

* Fixed device map

* Clippy

* Gemma3n no paged attn

* Fix saturating sub

* Faster rmsnorm

* Use sdpa for vision model

* Fix ple bug

* Fix name

* Fix multiaudio

* Add matformer config loading

* Add docs

* Add support for matformer in auto device mapper

* Update docs

* Typos

* Tweak

* Tweak

* Fix multidevice

* Fix gemma3n text model auto device map

* Fix dims3

* Fix auto devic emap vision

* Non-metal keeps PLE on cpu

* Complete merge

* Vision dtype f16 -> f32

* Fix metal nm device

* Fix uqff

* Typos

* Reference uqff

* Fix tests
2025-07-09 06:59:51 -04:00

6.8 KiB

1name# Layers# Effective Params (B)MMLU PT accuracyFFN Hidden DimsLayers Skipped
2Main model353.9862.30%[2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8]
3Config for official E2B Model301.9150.90%[2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4][20, 21, 22, 23, 24]
4Config for E1.96B (layer-level)301.9653.40%[2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 8, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4][20, 21, 22, 23, 24]
5Config for E2.54B (layer-level)352.5455.40%[2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 8, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4]
6Config for E2.69B (layer-level)352.6957.70%[2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 8, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 8, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 8, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4]
7Config for E2.98B (layer-level)352.9859.50%[2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8]
8Config for E3.18B (layer-level)353.1861.80%[2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 8, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 8, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 8]
9Config for E3.39B (layer-level)353.3963.00%[2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 8, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 8]
10Config for E3.59B (layer-level)353.5963.40%[2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 8]
11Config for E3.79B (layer-level)353.7963.40%[2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8]
12Config for E2.49B (block-level)352.4954.50%[2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4]
13Config for E2.73B (block-level)352.7357.10%[2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4]
14Config for E2.98B (block-level)352.9859.50%[2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4]
15Config for E3.24B (block-level)353.2460.70%[2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4]
16Config for E3.49B (block-level)353.4961.40%[2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4]
17Config for E3.79B (block-level)353.7462.00%[2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 4, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8, 2_048 * 8]