Upload README.md with huggingface_hub

Replace binaries from production build

Replace binaries from production build

Replace binaries from production build

Guides: porting Needle 3

8-29 MB everywhere

Replace binaries from production build

Replace binaries from production build

Tighter intro

Drop the redundant heading

Links under Get started instead of badges

Animated platform grid

One alternating fine-tuning chart

Animated model card and benchmark chart

Animated Cactus Needle 3 banner

Add the .cact format guide

Drop the Performance row

Badges under the banner

Minimal model card with guides and animated fine-tuning charts

Subnetwork wording, fine-tuning results

Fine-tuning results per rung

Upload README.md with huggingface_hub

Replace binaries from production build

Writing tools guidance, repair and withholding rules

Fill DeepSeek V4 Flash benchmark row

Benchmarks: drop the 2-layer rung

Benchmarks ordered by Mobile Actions

drop assets/architecture.png

drop assets/frontier.png

Needle 3 model card

checkpoints/needle3.safetensors: the 20-layer base (rungs are sliced by needle build --layers)

Upload README.md with huggingface_hub

Replace binaries from production build

checkpoints/needle3.safetensors: drop the speech vocabulary rows (vocab 8192)

Upload README.md with huggingface_hub

Replace binaries from production build

Add tokenizer.vocab

Add tokenizer.model

Add checkpoints/needle3.safetensors

Upload README.md with huggingface_hub

Replace binaries from production build

initial commit