# Choose models by task and memory

A v0.6.2 model hub for checkpoint selection, cache discovery, templates, tested Track G models, vision, sampling, and support boundaries.

## Direct answer

v0.6.2 adds checkpoint-tested Track G models including Seed-OSS-36B, with results that are checkpoint-specific, not family-wide performance guarantees. InternLM3-8B remains theoretical only because public checkpoints provide tokenizer.model while the Swift path requires tokenizer.json. Choose by exact architecture, tokenizer, template, memory, and task rather than assuming universal family support.

- Canonical: https://macmlx.app/models/
- Last verified: 2026-07-15

## Page facts

- **Apple Silicon unified memory** — Released; since 0.1.0; last verified 2026-07-15. 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.
- **Bounded model pool** — Released; since 0.5.0; last verified 2026-07-15. 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.
- **Supported LoRA adapters** — Released; since 0.5.0; last verified 2026-07-15. The native engine can apply supported LoRA adapters. Adapter compatibility depends on the base architecture and weights; universal LoRA compatibility is not claimed.
- **Fourteen detected VLM families** — Released; since 0.5.3; last verified 2026-07-15. 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.
- **DeepSeek V3.2 Swift overlay** — Released; since 0.5.3; last verified 2026-07-15. v0.5.3 includes pure-Swift component parity for the DeepSeek V3.2 architecture. A real-checkpoint smoke test remains pending and FP8 dequantization is absent, so this is not an end-to-end or universal MoE claim.
- **Speculative decoding** — Released; since 0.6.0; last verified 2026-07-15. v0.6.0 ships the classic draft-model path through both the API and GUI. Acceptance telemetry reports draft efficiency. Targets with non-trimmable hybrid or linear-attention caches are detected and fall back to standard decoding.
- **Hugging Face cache discovery** — Released; since 0.6.0; last verified 2026-07-15. 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.
- **Per-model chat-template overrides** — Released; since 0.6.2; last verified 2026-07-15. v0.6.2 resolves templates in user file, built-in model_type, then checkpoint order. Built-in overrides carry standard-path parity evidence, but template support is not universal; checkpoint-specific branches and unsupported swift-jinja syntax can still define the boundary.
- **Track G tested models** — Released; since 0.6.2; last verified 2026-07-15. v0.6.2 adds four checkpoint-tested native model families. Measured real-checkpoint generation: Seed-OSS-36B 4-bit at 18.2 tok/s; Hunyuan V1 Dense 1.8B 4-bit at 80.3 tok/s; Cohere Command R7B 7B 4-bit at 21.7 tok/s; and MiniCPM3-4B 4-bit at 18.7 tok/s. Results are checkpoint-specific, not family-wide performance guarantees.
- **InternLM3 theoretical support** — Theoretical; since 0.6.2; last verified 2026-07-15. v0.6.2 ships parity-verified InternLM3 code at the theoretical support tier. Real generation has not been demonstrated. Public checkpoints ship tokenizer.model but no tokenizer.json, while the Swift tokenizer stack requires tokenizer.json; support remains theoretical until that load-path boundary changes.
- **Temperature and top-p** — Released; since 0.1.0; last verified 2026-07-15. Temperature and nucleus top-p sampling are released controls. These are the current exposed core sampling controls.
- **Expanded sampling controls** — Planned; since future; last verified 2026-07-15. 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.

## Sources

- https://ml-explore.github.io/mlx/build/html/usage/unified_memory.html
- https://github.com/magicnight/mac-mlx/blob/v0.6.2/MacMLXCore/Sources/MacMLXCore/Engine/MLXSwiftEngine.swift
- https://github.com/magicnight/mac-mlx/releases/tag/v0.6.2
- https://github.com/magicnight/mac-mlx/blob/v0.6.2/CHANGELOG.md
- https://github.com/magicnight/mac-mlx/blob/v0.6.2/MacMLXCore/Sources/MacMLXCore/ModelPool/ModelPool.swift
- https://github.com/magicnight/mac-mlx/blob/v0.6.2/MacMLXCore/Sources/MacMLXCore/Managers/ModelLibraryManager.swift
- https://github.com/magicnight/mac-mlx/blob/v0.6.2/MacMLXCore/Sources/MacMLXCore/Models/DeepseekV32.swift
- https://github.com/magicnight/mac-mlx/blob/v0.6.2/MacMLXCore/Sources/MacMLXCore/Engine/SpeculativeDecodingUsage.swift
- https://github.com/magicnight/mac-mlx/blob/v0.6.2/MacMLXCore/Sources/MacMLXCore/Engine/ChatTemplateOverride.swift
- https://github.com/magicnight/mac-mlx/blob/v0.6.2/docs/model-support.md
- https://github.com/magicnight/mac-mlx/blob/v0.6.2/MacMLXCore/Sources/MacMLXCore/Models/InternLM3.swift
- https://github.com/magicnight/mac-mlx/blob/v0.6.2/MacMLXCore/Sources/MacMLXCore/Managers/ModelParametersStore.swift
- https://github.com/magicnight/mac-mlx/blob/main/docs/superpowers/specs/2026-07-10-engine-scroll-story-design.md
- https://github.com/magicnight/mac-mlx/blob/main/MacMLXCore/Sources/MacMLXCore/Managers/ModelParametersStore.swift

## Related pages

- [How to choose an MLX model](https://macmlx.app/models/choosing-a-model/)
- [Vision-language model support](https://macmlx.app/models/vision-language-models/)
- [How macMLX runs models on Apple Silicon](https://macmlx.app/architecture/)
- [macMLX questions, answered](https://macmlx.app/faq/)
- [macMLX and LM Studio](https://macmlx.app/compare/lm-studio/)