File size: 1,473 Bytes
498b819
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
7a9304a
498b819
2f72b7f
 
 
 
 
 
 
47c479e
 
2f72b7f
7a9304a
498b819
 
 
 
 
 
 
 
 
 
 
 
7a9304a
498b819
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
---
license: apache-2.0
license_link: https://huggingface.co/Qwen/Qwen2.5-Coder-1.5B-Instruct/blob/main/LICENSE
language:
- en
base_model: Qwen/Qwen2.5-Coder-1.5B-Instruct
pipeline_tag: text-generation
library_name: transformers
tags:
- code
- codeqwen
- chat
- qwen
- qwen-coder
- mlx
---

# bobig/Qwen2.5-Coder-1.5B-Instruct-Q6


This works well as a draft model for speculative decoding in [LMstudio 3.10 beta](https://lmstudio.ai/docs/advanced/speculative-decoding)

Try it with: [mlx-community/Qwen2.5-14B-1M-YOYO-V2-Q4](https://huggingface.co/mlx-community/Qwen2.5-14B-1M-YOYO-V2-Q4)

you should see about 50% faster TPS for math/code prompts. For a quick test try: "count backwards from 100 to 1"

Q4 was a little too dumb, Q8 was a little too slow...so Q6 


The Model [bobig/Qwen2.5-Coder-1.5B-Instruct-Q6](https://huggingface.co/bobig/Qwen2.5-Coder-1.5B-Instruct-Q6) was
converted to MLX format from [Qwen/Qwen2.5-Coder-1.5B-Instruct](https://huggingface.co/Qwen/Qwen2.5-Coder-1.5B-Instruct)
using mlx-lm version **0.21.4**.

## Use with mlx

```bash
pip install mlx-lm
```

```python
from mlx_lm import load, generate

model, tokenizer = load("bobig/Qwen2.5-Coder-1.5B-Instruct-Q6")

prompt = "hello"

if tokenizer.chat_template is not None:
    messages = [{"role": "user", "content": prompt}]
    prompt = tokenizer.apply_chat_template(
        messages, add_generation_prompt=True
    )

response = generate(model, tokenizer, prompt=prompt, verbose=True)
```