Sizing facts

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

Released0.5.0

Bounded model pool

Budgets, LRU eviction, pinning, cold swap, idle TTL, and probes bound multi-model use.

The pool shipped in v0.5.0 and was hardened in v0.5.3. It is not a unified adaptive controller.

Verified

Released0.5.0

Exact-prefix RAM and SSD cache

A hot RAM tier and content-addressed SSD cold tier support promotion and demotion.

The released v0.5.0 cache reuses exact full prefixes. It does not provide released block sharing or paged KV allocation.

Verified

Released0.6.0

KV-cache quantization

v0.6.0 exposes kv_bits, kv_group_size, and quantized_kv_start for compatible requests.

These controls change KV-cache precision, not model-weight precision. Nonpositive kv_bits disables the feature, and the compatibility matrix rejects unsupported model or request combinations.

Verified

Released0.6.0

Hugging Face cache discovery

v0.6.0 can discover models in configured Hugging Face cache roots without downloading duplicate weights.

Discovery confirms a local candidate, not a universal load guarantee; architecture, tokenizer, processor, and checkpoint compatibility still apply.

Verified

Released0.1.0

Temperature and top-p

Temperature and nucleus top-p sampling are released controls.

These are the current exposed core sampling controls.

Verified

Plannedfuture

Expanded sampling controls

top-k, min-p, presence, frequency, and repetition penalties, plus per-request seed are planned.

DeepSeek expert-routing top-k is an internal architecture operation and is unrelated to user sampling top-k.

Verified

A practical sizing sequence

Evidence-backed categories to check before downloading
FactorStart withWhy it matters
TaskText, code, embedding, or visionThe task determines the required architecture family
Parameter countThe smallest model that meets the taskWeights dominate baseline memory
QuantizationA supported 4-bit or 8-bit checkpointLower precision usually reduces weight memory
ContextThe shortest useful contextLonger context grows KV-cache pressure
ConcurrencyOne loaded model firstPools and simultaneous work add memory pressure
HeadroomLeave room for macOS and the workloadNear-capacity operation is less resilient

Official sources

  1. Apple Silicon unified memory
  2. KV-cache quantization
  3. Temperature and top-p
  4. Hugging Face cache discovery
  5. Bounded model pool
  6. Exact-prefix RAM and SSD cache
  7. KV-cache quantization
  8. Hugging Face cache discovery
  9. Temperature and top-p
  10. Expanded sampling controls
  11. Expanded sampling controls