Vision facts

Released0.5.3

Fourteen detected VLM families

The model library detects 14 vision-language model_type families.

This is an evidence-backed family count, not a guarantee that every checkpoint or processor variant will load.

Verified

Released0.1.0

Apple Silicon unified memory

MLX arrays use the Mac's shared CPU/GPU memory system.

Unified memory reduces explicit transfers between CPU orchestration and integrated-GPU compute, but model weights, activations, and KV cache still consume finite physical memory.

Verified

Released0.6.0

Eligibility-gated continuous batching

v0.6.0 batches only eligible dense-text requests under real concurrency, with automatic serial fallback.

The tagged 4-client benchmark measured 2.5–3.2× aggregate throughput. VLM, speculative decoding, Ollama, Anthropic, and embeddings remain serial.

Verified

Vision checkpoint checklist

The model library detects 14 VLM model_type families; each checkpoint still needs these checks
CategoryQuestionBoundary
ArchitectureIs the exact model_type among the detected families?A generic MLX label is not enough
ProcessorDoes the checkpoint include compatible image preprocessing?Processor variants can differ within a family
QuantizationIs this quantized checkpoint supported by its architecture?Support is not universal across weights
MemoryDo weights, image tokens, activations, and cache fit?Vision adds workload beyond text weights
Serving pathDoes the selected API path accept vision input?Batching is dense-text-only; VLM requests remain serial

Official sources

  1. Eligibility-gated continuous batching
  2. Fourteen detected VLM families
  3. Apple Silicon unified memory
  4. Apple Silicon unified memory
  5. Eligibility-gated continuous batching
  6. Eligibility-gated continuous batching
  7. Eligibility-gated continuous batching