---
license: mit
language: en
library_name: mlx
pipeline_tag: text-generation
base_model: deepgrove/maple-preview
tags:
- mlx
- causal-lm
- mixture-of-experts
- reasoning
- custom-code
- bf16
---

# Maple Preview — BF16 MLX conversion

This repository contains the full-precision BF16 MLX conversion of [`deepgrove/maple-preview`](https://huggingface.co/deepgrove/maple-preview).

## This MLX upload

- Approximately 40.5 GB for the complete local model directory (40,448,443,564 bytes).
- Converted directly from the original Hugging Face BF16 safetensors.
- Includes the official Maple model implementation as `maple.py`.
- Requires custom-code loading (`trust_remote_code=True`). In oMLX, enable **Trust Remote Code** for this model.

This is an MLX conversion for local inference on Apple Silicon. Please follow the base model's MIT license and usage terms.

## Recommended generation settings

Maple is a reasoning-heavy model and may spend a substantial part of its response budget thinking. For the OpenAI-compatible API or oMLX UI, start with:

```text
temperature:         1.0
top_p:               0.95
top_k:               40
min_p:               0.05
repetition_penalty:  1.0
max_tokens:          8192 or higher
max context:         131072 tokens (native model limit)
```

These sampler values match DeepGrove's Maple `llama.cpp` setup. The model declares a native 131,072-token context window and does not require RoPE/YARN scaling for that window. Actual usable context may be lower on systems constrained by KV-cache memory; do not assume that extending beyond 131,072 tokens is supported.

---

## Base model description

# Maple-Preview

**DeepGrove · 2026**

Today we introduce Maple-Preview, an open-source 20B-A1B ternary-weight reasoning LLM. Maple-Preview has SOTA reasoning for its weight class and is even competitive with larger models. It solves IMO-level problems and runs at 200+ tokens/sec on a Mac mini M4, 5–16× faster than efficient models like Gemma 4, Qwen3.5, and gpt-oss.

- 20B-A1B Model
- 218 tok/s M4 Mac mini
- 5.31 GB Checkpoint
- 131,072 Token context

![Maple-Preview speed and performance frontier](assets/01-speed-frontier.png)

> [!NOTE]
> The included Transformers implementation depends on Triton and FlashAttention
> and is intended for a compatible CUDA environment. The reported Apple Silicon
> result uses a separate on-device runtime.

## Architecture

Maple-Preview is a 20B-A1B reasoning model designed from the start for efficient on-device inference. It utilizes a 24-layer, 256-expert (8 active) configuration with 3:1 SWA-512:GA attention.

## Evaluation

On benchmarks, Maple-Preview sets a new point on the Pareto frontier for both memory-to-performance and speed-to-performance, demonstrating its strong reasoning capabilities. However, we note that this preview is focused primarily on raw reasoning and, as such, may underperform on agentic benchmarks. We intend to continue improving general performance through extended training before Maple's full release.

![Benchmark score comparison](assets/05-benchmark-scores-table.png)

Capability comparison using the dense output head across LCBv6, AIME 2026, HMMT 2026, and GPQA-D.

## Limitations

This preview received minimal post-training for agentic tasks and only
small-scale general reinforcement learning.

## License

Maple-Preview is released under the [MIT License](LICENSE).