legraphista commited on
Commit
a0b49d9
β€’
1 Parent(s): 59cefef

Upload README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +166 -0
README.md ADDED
@@ -0,0 +1,166 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: Qwen/Qwen2-0.5B
3
+ inference: false
4
+ language:
5
+ - en
6
+ library_name: gguf
7
+ license: apache-2.0
8
+ pipeline_tag: text-generation
9
+ quantized_by: legraphista
10
+ tags:
11
+ - pretrained
12
+ - quantized
13
+ - GGUF
14
+ - imatrix
15
+ - quantization
16
+ - imat
17
+ - imatrix
18
+ - static
19
+ - 16bit
20
+ - 8bit
21
+ - 6bit
22
+ - 5bit
23
+ - 4bit
24
+ - 3bit
25
+ - 2bit
26
+ - 1bit
27
+ ---
28
+
29
+ # Qwen2-0.5B-IMat-GGUF
30
+ _Llama.cpp imatrix quantization of Qwen/Qwen2-0.5B_
31
+
32
+ Original Model: [Qwen/Qwen2-0.5B](https://huggingface.co/Qwen/Qwen2-0.5B)
33
+ Original dtype: `BF16` (`bfloat16`)
34
+ Quantized by: llama.cpp [b3091](https://github.com/ggerganov/llama.cpp/releases/tag/b3091)
35
+ IMatrix dataset: [here](https://gist.githubusercontent.com/bartowski1182/eb213dccb3571f863da82e99418f81e8/raw/b2869d80f5c16fd7082594248e80144677736635/calibration_datav3.txt)
36
+
37
+ - [Files](#files)
38
+ - [IMatrix](#imatrix)
39
+ - [Common Quants](#common-quants)
40
+ - [All Quants](#all-quants)
41
+ - [Downloading using huggingface-cli](#downloading-using-huggingface-cli)
42
+ - [Inference](#inference)
43
+ - [Simple chat template](#simple-chat-template)
44
+ - [Chat template with system prompt](#chat-template-with-system-prompt)
45
+ - [Llama.cpp](#llama-cpp)
46
+ - [FAQ](#faq)
47
+ - [Why is the IMatrix not applied everywhere?](#why-is-the-imatrix-not-applied-everywhere)
48
+ - [How do I merge a split GGUF?](#how-do-i-merge-a-split-gguf)
49
+
50
+ ---
51
+
52
+ ## Files
53
+
54
+ ### IMatrix
55
+ Status: ⏳ Processing
56
+ Link: [here](https://huggingface.co/legraphista/Qwen2-0.5B-IMat-GGUF/blob/main/imatrix.dat)
57
+
58
+ ### Common Quants
59
+ | Filename | Quant type | File Size | Status | Uses IMatrix | Is Split |
60
+ | -------- | ---------- | --------- | ------ | ------------ | -------- |
61
+ | Qwen2-0.5B.Q8_0 | Q8_0 | - | ⏳ Processing | βšͺ Static | -
62
+ | Qwen2-0.5B.Q6_K | Q6_K | - | ⏳ Processing | βšͺ Static | -
63
+ | Qwen2-0.5B.Q4_K | Q4_K | - | ⏳ Processing | 🟒 IMatrix | -
64
+ | Qwen2-0.5B.Q3_K | Q3_K | - | ⏳ Processing | 🟒 IMatrix | -
65
+ | Qwen2-0.5B.Q2_K | Q2_K | - | ⏳ Processing | 🟒 IMatrix | -
66
+
67
+
68
+ ### All Quants
69
+ | Filename | Quant type | File Size | Status | Uses IMatrix | Is Split |
70
+ | -------- | ---------- | --------- | ------ | ------------ | -------- |
71
+ | Qwen2-0.5B.BF16 | BF16 | - | ⏳ Processing | βšͺ Static | -
72
+ | Qwen2-0.5B.FP16 | F16 | - | ⏳ Processing | βšͺ Static | -
73
+ | Qwen2-0.5B.Q8_0 | Q8_0 | - | ⏳ Processing | βšͺ Static | -
74
+ | Qwen2-0.5B.Q6_K | Q6_K | - | ⏳ Processing | βšͺ Static | -
75
+ | Qwen2-0.5B.Q5_K | Q5_K | - | ⏳ Processing | βšͺ Static | -
76
+ | Qwen2-0.5B.Q5_K_S | Q5_K_S | - | ⏳ Processing | βšͺ Static | -
77
+ | Qwen2-0.5B.Q4_K | Q4_K | - | ⏳ Processing | 🟒 IMatrix | -
78
+ | Qwen2-0.5B.Q4_K_S | Q4_K_S | - | ⏳ Processing | 🟒 IMatrix | -
79
+ | Qwen2-0.5B.IQ4_NL | IQ4_NL | - | ⏳ Processing | 🟒 IMatrix | -
80
+ | Qwen2-0.5B.IQ4_XS | IQ4_XS | - | ⏳ Processing | 🟒 IMatrix | -
81
+ | Qwen2-0.5B.Q3_K | Q3_K | - | ⏳ Processing | 🟒 IMatrix | -
82
+ | Qwen2-0.5B.Q3_K_L | Q3_K_L | - | ⏳ Processing | 🟒 IMatrix | -
83
+ | Qwen2-0.5B.Q3_K_S | Q3_K_S | - | ⏳ Processing | 🟒 IMatrix | -
84
+ | Qwen2-0.5B.IQ3_M | IQ3_M | - | ⏳ Processing | 🟒 IMatrix | -
85
+ | Qwen2-0.5B.IQ3_S | IQ3_S | - | ⏳ Processing | 🟒 IMatrix | -
86
+ | Qwen2-0.5B.IQ3_XS | IQ3_XS | - | ⏳ Processing | 🟒 IMatrix | -
87
+ | Qwen2-0.5B.IQ3_XXS | IQ3_XXS | - | ⏳ Processing | 🟒 IMatrix | -
88
+ | Qwen2-0.5B.Q2_K | Q2_K | - | ⏳ Processing | 🟒 IMatrix | -
89
+ | Qwen2-0.5B.Q2_K_S | Q2_K_S | - | ⏳ Processing | 🟒 IMatrix | -
90
+ | Qwen2-0.5B.IQ2_M | IQ2_M | - | ⏳ Processing | 🟒 IMatrix | -
91
+ | Qwen2-0.5B.IQ2_S | IQ2_S | - | ⏳ Processing | 🟒 IMatrix | -
92
+ | Qwen2-0.5B.IQ2_XS | IQ2_XS | - | ⏳ Processing | 🟒 IMatrix | -
93
+ | Qwen2-0.5B.IQ2_XXS | IQ2_XXS | - | ⏳ Processing | 🟒 IMatrix | -
94
+ | Qwen2-0.5B.IQ1_M | IQ1_M | - | ⏳ Processing | 🟒 IMatrix | -
95
+ | Qwen2-0.5B.IQ1_S | IQ1_S | - | ⏳ Processing | 🟒 IMatrix | -
96
+
97
+
98
+ ## Downloading using huggingface-cli
99
+ If you do not have hugginface-cli installed:
100
+ ```
101
+ pip install -U "huggingface_hub[cli]"
102
+ ```
103
+ Download the specific file you want:
104
+ ```
105
+ huggingface-cli download legraphista/Qwen2-0.5B-IMat-GGUF --include "Qwen2-0.5B.Q8_0.gguf" --local-dir ./
106
+ ```
107
+ If the model file is big, it has been split into multiple files. In order to download them all to a local folder, run:
108
+ ```
109
+ huggingface-cli download legraphista/Qwen2-0.5B-IMat-GGUF --include "Qwen2-0.5B.Q8_0/*" --local-dir ./
110
+ # see FAQ for merging GGUF's
111
+ ```
112
+
113
+ ---
114
+
115
+ ## Inference
116
+
117
+ ### Simple chat template
118
+ ```
119
+ <|im_start|>system
120
+ You are a helpful assistant<|im_end|>
121
+ <|im_start|>user
122
+ {user_prompt}<|im_end|>
123
+ <|im_start|>assistant
124
+ {assistant_response}<|im_end|>
125
+ <|im_start|>user
126
+ {next_user_prompt}<|im_end|>
127
+
128
+ ```
129
+
130
+ ### Chat template with system prompt
131
+ ```
132
+ <|im_start|>system
133
+ {system_prompt}<|im_end|>
134
+ <|im_start|>user
135
+ {user_prompt}<|im_end|>
136
+ <|im_start|>assistant
137
+ {assistant_response}<|im_end|>
138
+ <|im_start|>user
139
+ {next_user_prompt}<|im_end|>
140
+
141
+ ```
142
+
143
+ ### Llama.cpp
144
+ ```
145
+ llama.cpp/main -m Qwen2-0.5B.Q8_0.gguf --color -i -p "prompt here (according to the chat template)"
146
+ ```
147
+
148
+ ---
149
+
150
+ ## FAQ
151
+
152
+ ### Why is the IMatrix not applied everywhere?
153
+ 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).
154
+
155
+ ### How do I merge a split GGUF?
156
+ 1. Make sure you have `gguf-split` available
157
+ - To get hold of `gguf-split`, navigate to https://github.com/ggerganov/llama.cpp/releases
158
+ - Download the appropriate zip for your system from the latest release
159
+ - Unzip the archive and you should be able to find `gguf-split`
160
+ 2. Locate your GGUF chunks folder (ex: `Qwen2-0.5B.Q8_0`)
161
+ 3. Run `gguf-split --merge Qwen2-0.5B.Q8_0/Qwen2-0.5B.Q8_0-00001-of-XXXXX.gguf Qwen2-0.5B.Q8_0.gguf`
162
+ - Make sure to point `gguf-split` to the first chunk of the split.
163
+
164
+ ---
165
+
166
+ Got a suggestion? Ping me [@legraphista](https://x.com/legraphista)!