Alibaba Qwen · First MLX build 11 Nov 2024
Qwen2.5-Coder-32B-Instruct
MLX
Text apache-2.0
A 33B parameter text generation model from Alibaba Qwen, with 5 MLX builds, from 3-bit to bf16.
- Parameters
- 33B From Hugging Face
- Quantisation
- 4-bit Qwen2.5-Coder-32B-Instruct-4bit
- Memory needed
- 23 GB Estimate, unified memory
- Download
- 18.4 GB Weights, estimated
- Downloads
- 2.8K This build, last 30 days
- Task
- Text generation Text
- MLX builds
- 5 On mlx-community
- Devices
- Mac Apple silicon
Memory
Will it fit?Needs about 23 GB at 4-bit
8 GB
16 GB
24 GB
32 GB
64 GB
128 GB
Fits with headroomTight, close other appsWon’t fit
Estimated from the size of the weights, plus a fifth for context and working memory and 2 GB for macOS and your apps. Long contexts need more.
Run it
Terminal · mlx-lm$ pip install -U mlx-lm
$ mlx_lm.generate --model mlx-community/Qwen2.5-Coder-32B-Instruct-4bit --prompt "Hello"MLX builds
5 on mlx-community| Qwen2.5-Coder-32B-Instruct-3bit | 3-bit | 14.3 GB | 19 GB | 257 |
|---|---|---|---|---|
| Qwen2.5-Coder-32B-Instruct-4bit | 4-bit | 18.4 GB | 23 GB | 2.8K |
| Qwen2.5-Coder-32B-Instruct-6bit | 6-bit | 26.6 GB | 32 GB | 954 |
| Qwen2.5-Coder-32B-Instruct-8bit | 8-bit | 34.8 GB | 41 GB | 2.4K |
| Qwen2.5-Coder-32B-Instruct-bf16 | bf16 | 65.5 GB | 76 GB | 1.1K |
Sizes are estimated from Hugging Face’s parameter counts and the quantisation. Downloads are for the last 30 days.