aufklarer/Qwen3-ASR-1.7B-MLX-8bit

2B • Updated Apr 12 • 7.09k • 5

Note Default transcription. 8-bit MLX conversion of Qwen/Qwen3-ASR-1.7B (2.47 GB). Apache-2.0.

FluidInference/nemotron-3-diarization-coreml

Voice Activity Detection • Updated Sep 25 • 642 • 6

Note Default speaker diarization. Core ML offline model (0.2 GB) of nvidia/Nemotron-3-Diarization, run through FluidAudio. 14.6% DER vs 43.4% for the previous Pyannote setup on AMI. OpenMDW-1.1.

unsloth/Qwen3.5-4B-GGUF

Image-Text-to-Text • 4B • Updated Mar 2 • 1.52M • 549

Note Default for summaries and chat. Q4_K_M (2.74 GB): a 26-minute meeting to notes in 33 s warm, about 85 tok/s decode on M4 Max. Apache-2.0.

mradermacher/harrier-oss-v1-0.6b-GGUF

0.6B • Updated Mar 31 • 5.9k • 4

Note Default semantic search. Q8_0 (0.64 GB) on a dedicated llama-server with embeddings enabled. Correct passage first for 40/48 test queries vs 35/48 for Qwen3 Embedding 0.6B. MIT.

unsloth/LFM2.5-1.2B-Instruct-GGUF

Text Generation • 1B • Updated Aug 5 • 36.1k • 46

Note Default autocomplete. Q4_K_M (0.73 GB) of LiquidAI's model; 484 ms p95 in LokalBot's cotyping gate. LFM Open License: check revenue eligibility.