Views
No views yet
transformers or PyTorch. They use a custom layout produced by NNOpt and are consumed by the C++/OpenCL inference binaries in the GitHub repo above.1git clone https://github.com/a8nova/adreno-llms.git
2cd adreno-llms
3./scripts/fetch_weights.sh smollm2-135m-instruct # pulls from this repo
4cd src/models/smollm2-135m-instruct
5NNOPT_DTYPE=fp16 ./scripts/build.sh --release
6NNOPT_DTYPE=fp16 ./scripts/deploy_android.sh
7NNOPT_DTYPE=fp16 ./scripts/run_android.sh "Once upon a time" 64temperature=0, seed=42), 32-token generation, on Motorola Razr 2020 (Adreno 618), measured 2026-05-06.| Path | Upstream | Params | Decode tok/s | License of upstream weights |
|---|---|---|---|---|
mamba2-130m/model.fp16.bin | state-spaces/mamba2-130m | 130M | 23.18 | Apache 2.0 |
mamba-130m/model.fp16.bin | state-spaces/mamba-130m-hf | 130M | 22.15 | Apache 2.0 |
smollm2-135m-instruct/model.fp16.bin | HuggingFaceTB/SmolLM2-135M-Instruct | 135M | 14.57 | Apache 2.0 |
lfm2-5-350m-instruct/model.fp16.bin | LiquidAI/LFM2.5-350M | 350M | 10.20 | Liquid AI Open License |
qwen2-5-0-5b-instruct/model.fp16.bin | Qwen/Qwen2.5-0.5B-Instruct | 500M | 8.45 | Apache 2.0 |
whisper-tiny/model.fp16.bin | openai/whisper-tiny | 39M | ASR (see repo) | Apache 2.0 |
kokoro-82m/model.fp16.bin | hexgrad/Kokoro-82M | 82M | TTS (see repo) | Apache 2.0 |
musicgen-small/model.fp16.bin | facebook/musicgen-small | ~590M | text→music (see repo) | CC-BY-NC 4.0 |
seamless-m4t-unity-small/model.fp16.bin | facebook/seamless-m4t-unity-small | ~323M | S2ST/S2TT (see repo) | CC-BY-NC 4.0 |
openelm-270m-instruct/ (companion files only) | apple/OpenELM-270M-Instruct | 270M | 4.47 | Apple ASCL — fetch + convert locally |
openvoice-v2/model.fp16.bin | myshell-ai/OpenVoiceV2 | ~32M | voice cloning (see repo) | MIT |
pocket-tts/model.fp16.bin | kyutai/pocket-tts | ~100M | TTS (see repo) | see upstream (gated repo) |
moonshine-tiny/model.fp16.bin | UsefulSensors/moonshine-tiny | ~27M | streaming ASR (see repo) | MIT |
depth-anything-v2-small/model.fp16.bin | depth-anything/Depth-Anything-V2-Small-hf | ~25M | depth estimation (see repo) | Apache 2.0 |
stable-audio-open-small/model.fp16.bin | stabilityai/stable-audio-open-small | ~341M | text→audio (see repo) | Stability AI Community License |
openelm-270m-instruct/ you'll find only the small companion files:openelm-270m-instruct/model.fp16.meta.json # tensor layout for the C++ runtime
openelm-270m-instruct/tokenizer.json # HuggingFace tokenizer config
openelm-270m-instruct/tokenizer_vocab.bin # vocab + merges (binary)model.fp16.bin is NOT redistributed — Apple's Apple Sample Code License restricts that. Instead, scripts/fetch_openelm_weights.sh in the GitHub repo pulls apple/OpenELM-270M's safetensors directly from Apple's HF and runs scripts/convert_openelm_weights.py to produce the binary locally using the layout described in model.fp16.meta.json.