Issues

• Serve MiniCPM-o 4.5 — omni chat: text/image/audio in → text + 24 kHz speech out

  InferMatrixLab/vllm-omni-mlx#111 · Filter by author hsliuustc0106 opened on Oct 7, 2026
  1

• TTS model: CosyVoice3-0.5B (tier 2, watch) — 161k/mo, 9 languages, instruct + voice conversion; blocked on mlx-audio backend merge

  InferMatrixLab/vllm-omni-mlx#105 · Filter by author hsliuustc0106 opened on Oct 6, 2026

• Image model: Qwen-Image-2.1 + fast variants (tier 1) — the currently-trending image family (3M+/mo wave), upstream flagship parity

  InferMatrixLab/vllm-omni-mlx#101 · Filter by author hsliuustc0106 opened on Oct 6, 2026
  1

• Image model: FLUX.2-klein-4B (tier 1) — 394k/mo, 4-step, Apache-2.0, smallest of the category, edit-native

  InferMatrixLab/vllm-omni-mlx#100 · Filter by author hsliuustc0106 opened on Oct 6, 2026

• Image model: Z-Image-Turbo 6B (tier 1, engine reference path candidate) — 592k/mo, 9-step, Apache-2.0, mflux-native

  InferMatrixLab/vllm-omni-mlx#99 · Filter by author hsliuustc0106 opened on Oct 6, 2026
  1

• Image fast lane: low-step generation — SDXL-Turbo (943k/mo) vs SDXL-Lightning vs LCM-LoRA

  InferMatrixLab/vllm-omni-mlx#97 · Filter by author hsliuustc0106 opened on Oct 6, 2026

• Image model: Stable Diffusion 1.5 4-bit (tier 1) — 2nd-hottest image model on the Hub, ~1 GB, small+hot pick

  InferMatrixLab/vllm-omni-mlx#96 · Filter by author hsliuustc0106 opened on Oct 6, 2026

• Image diffusion breadth: tiered model support list — engine + FLUX.1-schnell (mflux), SDXL 4-bit, SD3.5-medium, (watch) Qwen-Image

  InferMatrixLab/vllm-omni-mlx#95 · Filter by author hsliuustc0106 opened on Oct 6, 2026
  3

• Image model: SD3.5-medium 4-bit (tier 2) — MMDiT tier, upstream's images-API flagship

  InferMatrixLab/vllm-omni-mlx#94 · Filter by author hsliuustc0106 opened on Oct 6, 2026

• Image model: SDXL-base 4-bit (tier 1) — most-used image model on the Hub, CLIP-only memory tier

  InferMatrixLab/vllm-omni-mlx#93 · Filter by author hsliuustc0106 opened on Oct 6, 2026

• Image model: FLUX.1-schnell 4-bit (tier 1) — mflux seam, 4-step fast lane, txt2img + img2img

  InferMatrixLab/vllm-omni-mlx#92 · Filter by author hsliuustc0106 opened on Oct 6, 2026

• Image generation engine: /v1/images/generations on MLX — mflux seam, [image] extra, time-to-image doctrine

  InferMatrixLab/vllm-omni-mlx#91 · Filter by author hsliuustc0106 opened on Oct 6, 2026
  4

• Reproduce the quiet-machine TTS streaming benchmarks on your Apple Silicon (RTF record: 0.37–0.42)

  InferMatrixLab/vllm-omni-mlx#84 · Filter by author hsliuustc0106 opened on Oct 5, 2026
  4

• TTS model: Voxtral-4B-TTS-2603 (tier 2, watch) — AR + FlowMatching presets

  InferMatrixLab/vllm-omni-mlx#74 · Filter by author hsliuustc0106 opened on Oct 5, 2026

• TTS model: MOSS-TTS-Nano-100M (tier 2) — tiny single-stage AR, cloning-only

  InferMatrixLab/vllm-omni-mlx#73 · Filter by author hsliuustc0106 opened on Oct 5, 2026
  1

  

• TTS model: OmniVoice (tier 1) — diffusion cloning tier, bf16 1.2 GB

  InferMatrixLab/vllm-omni-mlx#72 · Filter by author hsliuustc0106 opened on Oct 5, 2026

• TTS model: Kokoro-82M (tier 1) — non-AR fast lane, preset voices, streaming by sentence chunking

  InferMatrixLab/vllm-omni-mlx#70 · Filter by author hsliuustc0106 opened on Oct 5, 2026
  1

• TTS breadth: tiered model support list — Kokoro-82M, VoxCPM2, OmniVoice, MOSS-TTS-Nano, (watch) Voxtral-4B-TTS

  InferMatrixLab/vllm-omni-mlx#69 · Filter by author hsliuustc0106 opened on Oct 5, 2026

• ASR decoder-style first: /v1/audio/transcriptions on qwen3_asr — prompt-prefix cache + TPOT/ITL doctrine

  InferMatrixLab/vllm-omni-mlx#68 · Filter by author hsliuustc0106 opened on Oct 5, 2026
  2

• Roadmap: batch-1 low-latency omni serving on Apple Silicon — v0.1 core + M1 TTS

  InferMatrixLab/vllm-omni-mlx#2 · Filter by author hsliuustc0106 opened on Oct 2, 2026
  11