🤝 Open to Collab

Anurag

edwixx

27 40 163

https://anuragkanade.com

AI & ML interests

ASR, TTS, Speech and Audio

Recent Activity

liked a model 32 minutes ago

IndexTeam/IndexTTS-2.5

liked a model 1 day ago

Lightricks/LTX-2.5

liked a model 2 days ago

KRAFTON/Raon-Speech-9B

View all activity

Organizations

liked a model 32 minutes ago

IndexTeam/IndexTTS-2.5

Text-to-Speech • Updated about 14 hours ago • 58

liked a model 1 day ago

Lightricks/LTX-2.5

Image-to-Video • Updated about 7 hours ago • 39 • 549

liked 2 models 2 days ago

KRAFTON/Raon-Speech-9B

Any-to-Any • 9B • Updated Jul 7 • 5.31k • 54

meta-models/Muse-Glimmer-30B

Image-Text-to-Text • 30B • Updated 1 day ago • • 1.27k

published a bucket 4 days ago

edwixx/yt-music-temp

0 Bytes

liked a model 5 days ago

nvidia/NVIDIA-NemotronLabs-VoiceChat-11B

11B • Updated 1 day ago • 653 • 349

liked a model 11 days ago

neuphonic/neutts-2e

Text-to-Speech • 0.2B • Updated 21 days ago • 8.55k • 40

liked a model 12 days ago

MIT/ast-finetuned-audioset-10-10-0.4593

Audio Classification • 86.6M • Updated Sep 6, 2023 • 1.24M • 364

liked 2 models 13 days ago

pyannote/overlapped-speech-detection

Automatic Speech Recognition • Updated May 10, 2024 • 24.2k • 63

pyannote/brouhaha

Voice Activity Detection • Updated Nov 15, 2022 • 24k • 30

updated a bucket 14 days ago

edwixx/s2s-a0-en

315 MB

published a bucket 16 days ago

edwixx/s2s-a0-en

315 MB

liked a dataset 16 days ago

datacurve/deep-swe

Benchmark • Updated Jun 2 • 113 • 695 • 56

New activity in moonshotai/Kimi-K3 16 days ago

SOTA

🚀🤗 27

#39 opened 16 days ago by

edwixx

liked a model 16 days ago

moonshotai/Kimi-K3

Image-Text-to-Text • 2.8T • Updated 16 days ago • 1.57M • • 10.6k

updated a dataset 19 days ago

edwixx/s2s-blog-assets

Viewer • Updated 19 days ago • 1 • 871

liked a dataset 19 days ago

hiraki/candor-turntaking-annotations

Viewer • Updated Mar 25 • 173k • 95 • 1

reacted to SeaWolf-AI's post with 🔥 21 days ago

Post

4051

📱 POCKET — a 35-billion-parameter model that runs on your iPhone, and on your PC with no GPU

We're releasing POCKET, VIDRAFT's flagship Darwin-36B-Opus compressed for on-device use. No fork, no CUDA, no cloud — it runs on stock llama.cpp. It's a sparse Mixture-of-Experts model (256 experts, only 8 active per token), so the file can be large while the work per token stays small. That's what lets a 35B model run on a phone, and generate fast on a CPU with no graphics card.

Measured (POCKET-35B IQ1_M vs Bonsai-27B Q1_0):
• CPU generate (Xeon, 16 threads): 27.0 vs 10.1 tok/s → 2.69× faster
• GPU generate (H100): 197 vs 89 tok/s → 2.22× faster
• GPU prompt processing (H100): 753 vs 1816 → 0.41× (Bonsai wins this one — MoE prefill wakes every expert, so sparsity stops helping there. We say so.)
• Quality (HellaSwag, 400 q): 61.0% vs 60.0% → a tie (confidence intervals overlap)

On a real consumer laptop — MacBook M3 Pro (18 GB) — POCKET wins every axis, prompt processing included:
• Metal generate: 25.4 vs 12.8 → 1.99×
• CPU generate: 13.8 vs 4.4 → 3.13×
• Metal prompt: 240.7 vs 73.4 → 3.28×

One more quiet fact: the same-size, quality-oriented rival Ternary-Bonsai-27B (7.2 GB) fails to load in upstream llama.cpp at all — it needs the PrismML fork. POCKET runs on the tools you already have: LM Studio, Ollama, PocketPal, MLX.

📖 Full story (tech, measurements, recipes): https://huggingface.co/blog/FINAL-Bench/pocket

Models:
📦 POCKET-35B-GGUF (PC / server, no GPU): FINAL-Bench/POCKET-35B-GGUF
🇰🇷 POCKET-KR-GGUF (Android): FINAL-Bench/POCKET-KR-GGUF
🍎 POCKET-KR-MLX (iPhone / Mac): FINAL-Bench/POCKET-KR-MLX
🌍 POCKET-EN-GGUF (English phone / PC): FINAL-Bench/POCKET-EN-GGUF
🖥️ Live demo (answering on a CPU, no GPU): FINAL-Bench/POCKET-35B-CPU
📚 Collection: FINAL-Bench/pocket-models-6a618ee5d23eafb7e185a5c6