Hugging Face's logo

legraphista
/
Qwen2-72B-Instruct-IMat-GGUF

Text Generation
GGUF
English
chat
quantized
GGUF
imatrix
quantization
imat
static
16bit
8bit
6bit
5bit
4bit
3bit
2bit
1bit
conversational
Model card Files Community
legraphista commited on
Commit
533a9c7
•
1 Parent(s): 46c511e

Upload README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +168 -0
README.md ADDED
@@ -0,0 +1,168 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: Qwen/Qwen2-72B-Instruct
3
+ inference: false
4
+ language:
5
+ - en
6
+ library_name: gguf
7
+ license: other
8
+ license_link: https://huggingface.co/Qwen/Qwen2-72B-Instruct/blob/main/LICENSE
9
+ license_name: tongyi-qianwen
10
+ pipeline_tag: text-generation
11
+ quantized_by: legraphista
12
+ tags:
13
+ - chat
14
+ - quantized
15
+ - GGUF
16
+ - imatrix
17
+ - quantization
18
+ - imat
19
+ - imatrix
20
+ - static
21
+ - 16bit
22
+ - 8bit
23
+ - 6bit
24
+ - 5bit
25
+ - 4bit
26
+ - 3bit
27
+ - 2bit
28
+ - 1bit
29
+ ---
30
+
31
+ # Qwen2-72B-Instruct-IMat-GGUF
32
+ _Llama.cpp imatrix quantization of Qwen/Qwen2-72B-Instruct_
33
+
34
+ Original Model: [Qwen/Qwen2-72B-Instruct](https://huggingface.co/Qwen/Qwen2-72B-Instruct)
35
+ Original dtype: `BF16` (`bfloat16`)
36
+ Quantized by: llama.cpp [b3091](https://github.com/ggerganov/llama.cpp/releases/tag/b3091)
37
+ IMatrix dataset: [here](https://gist.githubusercontent.com/bartowski1182/eb213dccb3571f863da82e99418f81e8/raw/b2869d80f5c16fd7082594248e80144677736635/calibration_datav3.txt)
38
+
39
+ - [Files](#files)
40
+ - [IMatrix](#imatrix)
41
+ - [Common Quants](#common-quants)
42
+ - [All Quants](#all-quants)
43
+ - [Downloading using huggingface-cli](#downloading-using-huggingface-cli)
44
+ - [Inference](#inference)
45
+ - [Simple chat template](#simple-chat-template)
46
+ - [Chat template with system prompt](#chat-template-with-system-prompt)
47
+ - [Llama.cpp](#llama-cpp)
48
+ - [FAQ](#faq)
49
+ - [Why is the IMatrix not applied everywhere?](#why-is-the-imatrix-not-applied-everywhere)
50
+ - [How do I merge a split GGUF?](#how-do-i-merge-a-split-gguf)
51
+
52
+ ---
53
+
54
+ ## Files
55
+
56
+ ### IMatrix
57
+ Status: ⏳ Processing
58
+ Link: [here](https://huggingface.co/legraphista/Qwen2-72B-Instruct-IMat-GGUF/blob/main/imatrix.dat)
59
+
60
+ ### Common Quants
61
+ | Filename | Quant type | File Size | Status | Uses IMatrix | Is Split |
62
+ | -------- | ---------- | --------- | ------ | ------------ | -------- |
63
+ | Qwen2-72B-Instruct.Q8_0 | Q8_0 | - | ⏳ Processing | ⚪ Static | -
64
+ | Qwen2-72B-Instruct.Q6_K | Q6_K | - | ⏳ Processing | ⚪ Static | -
65
+ | Qwen2-72B-Instruct.Q4_K | Q4_K | - | ⏳ Processing | 🟢 IMatrix | -
66
+ | Qwen2-72B-Instruct.Q3_K | Q3_K | - | ⏳ Processing | 🟢 IMatrix | -
67
+ | Qwen2-72B-Instruct.Q2_K | Q2_K | - | ⏳ Processing | 🟢 IMatrix | -
68
+
69
+
70
+ ### All Quants
71
+ | Filename | Quant type | File Size | Status | Uses IMatrix | Is Split |
72
+ | -------- | ---------- | --------- | ------ | ------------ | -------- |
73
+ | Qwen2-72B-Instruct.BF16 | BF16 | - | ⏳ Processing | ⚪ Static | -
74
+ | Qwen2-72B-Instruct.FP16 | F16 | - | ⏳ Processing | ⚪ Static | -
75
+ | Qwen2-72B-Instruct.Q8_0 | Q8_0 | - | ⏳ Processing | ⚪ Static | -
76
+ | Qwen2-72B-Instruct.Q6_K | Q6_K | - | ⏳ Processing | ⚪ Static | -
77
+ | Qwen2-72B-Instruct.Q5_K | Q5_K | - | ⏳ Processing | ⚪ Static | -
78
+ | Qwen2-72B-Instruct.Q5_K_S | Q5_K_S | - | ⏳ Processing | ⚪ Static | -
79
+ | Qwen2-72B-Instruct.Q4_K | Q4_K | - | ⏳ Processing | 🟢 IMatrix | -
80
+ | Qwen2-72B-Instruct.Q4_K_S | Q4_K_S | - | ⏳ Processing | 🟢 IMatrix | -
81
+ | Qwen2-72B-Instruct.IQ4_NL | IQ4_NL | - | ⏳ Processing | 🟢 IMatrix | -
82
+ | Qwen2-72B-Instruct.IQ4_XS | IQ4_XS | - | ⏳ Processing | 🟢 IMatrix | -
83
+ | Qwen2-72B-Instruct.Q3_K | Q3_K | - | ⏳ Processing | 🟢 IMatrix | -
84
+ | Qwen2-72B-Instruct.Q3_K_L | Q3_K_L | - | ⏳ Processing | 🟢 IMatrix | -
85
+ | Qwen2-72B-Instruct.Q3_K_S | Q3_K_S | - | ⏳ Processing | 🟢 IMatrix | -
86
+ | Qwen2-72B-Instruct.IQ3_M | IQ3_M | - | ⏳ Processing | 🟢 IMatrix | -
87
+ | Qwen2-72B-Instruct.IQ3_S | IQ3_S | - | ⏳ Processing | 🟢 IMatrix | -
88
+ | Qwen2-72B-Instruct.IQ3_XS | IQ3_XS | - | ⏳ Processing | 🟢 IMatrix | -
89
+ | Qwen2-72B-Instruct.IQ3_XXS | IQ3_XXS | - | ⏳ Processing | 🟢 IMatrix | -
90
+ | Qwen2-72B-Instruct.Q2_K | Q2_K | - | ⏳ Processing | 🟢 IMatrix | -
91
+ | Qwen2-72B-Instruct.Q2_K_S | Q2_K_S | - | ⏳ Processing | 🟢 IMatrix | -
92
+ | Qwen2-72B-Instruct.IQ2_M | IQ2_M | - | ⏳ Processing | 🟢 IMatrix | -
93
+ | Qwen2-72B-Instruct.IQ2_S | IQ2_S | - | ⏳ Processing | 🟢 IMatrix | -
94
+ | Qwen2-72B-Instruct.IQ2_XS | IQ2_XS | - | ⏳ Processing | 🟢 IMatrix | -
95
+ | Qwen2-72B-Instruct.IQ2_XXS | IQ2_XXS | - | ⏳ Processing | 🟢 IMatrix | -
96
+ | Qwen2-72B-Instruct.IQ1_M | IQ1_M | - | ⏳ Processing | 🟢 IMatrix | -
97
+ | Qwen2-72B-Instruct.IQ1_S | IQ1_S | - | ⏳ Processing | 🟢 IMatrix | -
98
+
99
+
100
+ ## Downloading using huggingface-cli
101
+ If you do not have hugginface-cli installed:
102
+ ```
103
+ pip install -U "huggingface_hub[cli]"
104
+ ```
105
+ Download the specific file you want:
106
+ ```
107
+ huggingface-cli download legraphista/Qwen2-72B-Instruct-IMat-GGUF --include "Qwen2-72B-Instruct.Q8_0.gguf" --local-dir ./
108
+ ```
109
+ If the model file is big, it has been split into multiple files. In order to download them all to a local folder, run:
110
+ ```
111
+ huggingface-cli download legraphista/Qwen2-72B-Instruct-IMat-GGUF --include "Qwen2-72B-Instruct.Q8_0/*" --local-dir ./
112
+ # see FAQ for merging GGUF's
113
+ ```
114
+
115
+ ---
116
+
117
+ ## Inference
118
+
119
+ ### Simple chat template
120
+ ```
121
+ <|im_start|>system
122
+ You are a helpful assistant.<|im_end|>
123
+ <|im_start|>user
124
+ {user_prompt}<|im_end|>
125
+ <|im_start|>assistant
126
+ {assistant_response}<|im_end|>
127
+ <|im_start|>user
128
+ {next_user_prompt}<|im_end|>
129
+
130
+ ```
131
+
132
+ ### Chat template with system prompt
133
+ ```
134
+ <|im_start|>system
135
+ {system_prompt}<|im_end|>
136
+ <|im_start|>user
137
+ {user_prompt}<|im_end|>
138
+ <|im_start|>assistant
139
+ {assistant_response}<|im_end|>
140
+ <|im_start|>user
141
+ {next_user_prompt}<|im_end|>
142
+
143
+ ```
144
+
145
+ ### Llama.cpp
146
+ ```
147
+ llama.cpp/main -m Qwen2-72B-Instruct.Q8_0.gguf --color -i -p "prompt here (according to the chat template)"
148
+ ```
149
+
150
+ ---
151
+
152
+ ## FAQ
153
+
154
+ ### Why is the IMatrix not applied everywhere?
155
+ According to [this investigation](https://www.reddit.com/r/LocalLLaMA/comments/1993iro/ggufs_quants_can_punch_above_their_weights_now/), it appears that lower quantizations are the only ones that benefit from the imatrix input (as per hellaswag results).
156
+
157
+ ### How do I merge a split GGUF?
158
+ 1. Make sure you have `gguf-split` available
159
+ - To get hold of `gguf-split`, navigate to https://github.com/ggerganov/llama.cpp/releases
160
+ - Download the appropriate zip for your system from the latest release
161
+ - Unzip the archive and you should be able to find `gguf-split`
162
+ 2. Locate your GGUF chunks folder (ex: `Qwen2-72B-Instruct.Q8_0`)
163
+ 3. Run `gguf-split --merge Qwen2-72B-Instruct.Q8_0/Qwen2-72B-Instruct.Q8_0-00001-of-XXXXX.gguf Qwen2-72B-Instruct.Q8_0.gguf`
164
+ - Make sure to point `gguf-split` to the first chunk of the split.
165
+
166
+ ---
167
+
168
+ Got a suggestion? Ping me [@legraphista](https://x.com/legraphista)!