File size: 4,720 Bytes
e0bc976
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f31870a
e0bc976
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
789741e
e0bc976
 
 
 
 
 
 
 
 
 
 
6f9297d
 
 
 
e0bc976
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4f14824
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
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
---
base_model: Infinigence/Megrez-3B-Instruct
license: apache-2.0
model_creator: Infinigence
model_name: Megrez-3B-Instruct
quantized_by: Second State Inc.
language:
- en
- zh
pipeline_tag: text-generation
---

<!-- header start -->
<!-- 200823 -->
<div style="width: auto; margin-left: auto; margin-right: auto">
<img src="https://github.com/LlamaEdge/LlamaEdge/raw/dev/assets/logo.svg" style="width: 100%; min-width: 400px; display: block; margin: auto;">
</div>
<hr style="margin-top: 1.0em; margin-bottom: 1.0em;">
<!-- header end -->

# Megrez-3B-Instruct-GGUF

## Original Model

[Infinigence/Megrez-3B-Instruct](https://huggingface.co/Infinigence/Megrez-3B-Instruct)

## Run with LlamaEdge

- LlamaEdge version: [v0.16.0](https://github.com/LlamaEdge/LlamaEdge/releases/tag/0.16.0)

- Prompt template

  - Prompt type: `megrez`

  - Prompt string

    ```text
    <|role_start|>system<|role_end|>{system_message}<|turn_end|><|role_start|>user<|role_end|>{user_message}<|turn_end|><|role_start|>assistant<|role_end|>
    ```

- Context size: `32000`

- Run as LlamaEdge service

  ```bash
  wasmedge --dir .:. --nn-preload default:GGML:AUTO:Megrez-3B-Instruct-Q5_K_M.gguf \
    llama-api-server.wasm \
    --model-name Megrez-3B-Instruct \
    --prompt-template megrez \
    --ctx-size 32000
  ```

  > [!TIP]
  > For use cases of conversations or article writing, `temperature=0.7` is strongly recommended.
  > For use cases of mathematics or logical reasoning, `temperature=0.2` is strongly recommended.

- Run as LlamaEdge command app

  ```bash
  wasmedge --dir .:. --nn-preload default:GGML:AUTO:Megrez-3B-Instruct-Q5_K_M.gguf \
    llama-chat.wasm \
    --prompt-template megrez \
    --ctx-size 32000
  ```
  
  > [!TIP]
  > For use cases of conversations or article writing, `temperature=0.7` is strongly recommended.
  > For use cases of mathematics or logical reasoning, `temperature=0.2` is strongly recommended.

## Quantized GGUF Models

| Name | Quant method | Bits | Size | Use case |
| ---- | ---- | ---- | ---- | ----- |
| [Megrez-3B-Instruct-Q2_K.gguf](https://huggingface.co/second-state/Megrez-3B-Instruct-GGUF/blob/main/Megrez-3B-Instruct-Q2_K.gguf)     | Q2_K   | 2 | 1.21 GB| smallest, significant quality loss - not recommended for most purposes |
| [Megrez-3B-Instruct-Q3_K_L.gguf](https://huggingface.co/second-state/Megrez-3B-Instruct-GGUF/blob/main/Megrez-3B-Instruct-Q3_K_L.gguf) | Q3_K_L | 3 | 1.60 GB| small, substantial quality loss |
| [Megrez-3B-Instruct-Q3_K_M.gguf](https://huggingface.co/second-state/Megrez-3B-Instruct-GGUF/blob/main/Megrez-3B-Instruct-Q3_K_M.gguf) | Q3_K_M | 3 | 1.50 GB| very small, high quality loss |
| [Megrez-3B-Instruct-Q3_K_S.gguf](https://huggingface.co/second-state/Megrez-3B-Instruct-GGUF/blob/main/Megrez-3B-Instruct-Q3_K_S.gguf) | Q3_K_S | 3 | 1.38 GB| very small, high quality loss |
| [Megrez-3B-Instruct-Q4_0.gguf](https://huggingface.co/second-state/Megrez-3B-Instruct-GGUF/blob/main/Megrez-3B-Instruct-Q4_0.gguf)     | Q4_0   | 4 | 1.73 GB| legacy; small, very high quality loss - prefer using Q3_K_M |
| [Megrez-3B-Instruct-Q4_K_M.gguf](https://huggingface.co/second-state/Megrez-3B-Instruct-GGUF/blob/main/Megrez-3B-Instruct-Q4_K_M.gguf) | Q4_K_M | 4 | 1.81 GB| medium, balanced quality - recommended |
| [Megrez-3B-Instruct-Q4_K_S.gguf](https://huggingface.co/second-state/Megrez-3B-Instruct-GGUF/blob/main/Megrez-3B-Instruct-Q4_K_S.gguf) | Q4_K_S | 4 | 1.74 GB| small, greater quality loss |
| [Megrez-3B-Instruct-Q5_0.gguf](https://huggingface.co/second-state/Megrez-3B-Instruct-GGUF/blob/main/Megrez-3B-Instruct-Q5_0.gguf)     | Q5_0   | 5 | 2.05 GB| legacy; medium, balanced quality - prefer using Q4_K_M |
| [Megrez-3B-Instruct-Q5_K_M.gguf](https://huggingface.co/second-state/Megrez-3B-Instruct-GGUF/blob/main/Megrez-3B-Instruct-Q5_K_M.gguf) | Q5_K_M | 5 | 2.09 GB| large, very low quality loss - recommended |
| [Megrez-3B-Instruct-Q5_K_S.gguf](https://huggingface.co/second-state/Megrez-3B-Instruct-GGUF/blob/main/Megrez-3B-Instruct-Q5_K_S.gguf) | Q5_K_S | 5 | 2.05 GB| large, low quality loss - recommended |
| [Megrez-3B-Instruct-Q6_K.gguf](https://huggingface.co/second-state/Megrez-3B-Instruct-GGUF/blob/main/Megrez-3B-Instruct-Q6_K.gguf)     | Q6_K   | 6 | 2.40 GB| very large, extremely low quality loss |
| [Megrez-3B-Instruct-Q8_0.gguf](https://huggingface.co/second-state/Megrez-3B-Instruct-GGUF/blob/main/Megrez-3B-Instruct-Q8_0.gguf)     | Q8_0   | 8 | 3.10 GB| very large, extremely low quality loss - not recommended |
| [Megrez-3B-Instruct-f16.gguf](https://huggingface.co/second-state/Megrez-3B-Instruct-GGUF/blob/main/Megrez-3B-Instruct-f16.gguf)       | f16    | 16 | 5.84 GB| |

*Quantized with llama.cpp b4381*