---
title: maple-preview-webgpu
canonical_url: "https://www.modelscope.cn/models/ProCreations/maple-preview-webgpu"
md_url: "https://www.modelscope.cn/models/ProCreations/maple-preview-webgpu.md"
repository: ProCreations/maple-preview-webgpu
last_updated: 2026-08-05
license: mit
model_type:
  - maple
architectures:
  - MapleForCausalLM
base_model:
  - deepgrove/maple-preview
base_model_relation: quantized
library_name:
  - pytorch
frameworks:
  - pytorch
downloads: 0
stars: 0
tags:
  - webgpu
  - browser
  - text-generation
  - quantized
---

# maple-preview-webgpu

> maple-preview-webgpu - ProCreations 在 ModelScope 开源的模型。Maple-Preview WebGPU Pack

ProCreations/maple-preview-webgpu 是 ModelScope 魔搭社区上的机器学习模型，采用 mit 许可，基于 deepgrove/maple-preview 构建。

- **Repository**: ProCreations/maple-preview-webgpu
- **License**: mit
- **Base model**: deepgrove/maple-preview
- **Tags**: webgpu, browser, text-generation, quantized
- **Downloads**: 0
- **Stars**: 0
- **Last updated**: 2026-08-05

Source: https://www.modelscope.cn/models/ProCreations/maple-preview-webgpu

---

# Maple-Preview WebGPU Pack

Lossless browser-streaming repack of
[`deepgrove/maple-preview-2bit-mlx`](https://huggingface.co/deepgrove/maple-preview-2bit-mlx)
for [`ProCreations/maple-webgpu`](https://huggingface.co/spaces/ProCreations/maple-webgpu).

No model values are changed and no additional quantization is applied. The
official 2-bit/4-bit MLX tensors are regrouped into one aligned binary file per
transformer layer, reducing browser loading from hundreds of range requests to
a small set of sequential CDN downloads.

- Source revision: `361db5da5e74ff6fcdd852d478e1f266ce11013a`
- Packed tensor bytes: `5.31 GB`
- Container format: `maple-webgpu-pack-v1`
- Original model and weights: MIT, © DeepGrove AI

This is a runtime artifact, not a separate model release. See `manifest.json`
for exact byte offsets, shapes, dtypes, and source tensor names.