legraphista commited on
Commit
1d0b07a
β€’
1 Parent(s): 847bee2

Upload README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +169 -0
README.md ADDED
@@ -0,0 +1,169 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: failspy/Llama-3-8B-Instruct-MopeyMule
3
+ inference: false
4
+ language:
5
+ - en
6
+ library_name: gguf
7
+ license: other
8
+ license_name: llama3
9
+ pipeline_tag: text-generation
10
+ quantized_by: legraphista
11
+ tags:
12
+ - facebook
13
+ - meta
14
+ - pytorch
15
+ - llama
16
+ - llama-3
17
+ - quantized
18
+ - GGUF
19
+ - imatrix
20
+ - quantization
21
+ - imat
22
+ - imatrix
23
+ - static
24
+ - 16bit
25
+ - 8bit
26
+ - 6bit
27
+ - 5bit
28
+ - 4bit
29
+ - 3bit
30
+ - 2bit
31
+ - 1bit
32
+ ---
33
+
34
+ # Llama-3-8B-Instruct-MopeyMule-IMat-GGUF
35
+ _Llama.cpp imatrix quantization of failspy/Llama-3-8B-Instruct-MopeyMule_
36
+
37
+ Original Model: [failspy/Llama-3-8B-Instruct-MopeyMule](https://huggingface.co/failspy/Llama-3-8B-Instruct-MopeyMule)
38
+ Original dtype: `BF16` (`bfloat16`)
39
+ Quantized by: llama.cpp [b3056](https://github.com/ggerganov/llama.cpp/releases/tag/b3056)
40
+ IMatrix dataset: [here](https://gist.githubusercontent.com/bartowski1182/eb213dccb3571f863da82e99418f81e8/raw/b2869d80f5c16fd7082594248e80144677736635/calibration_datav3.txt)
41
+
42
+ - [Files](#files)
43
+ - [IMatrix](#imatrix)
44
+ - [Common Quants](#common-quants)
45
+ - [All Quants](#all-quants)
46
+ - [Downloading using huggingface-cli](#downloading-using-huggingface-cli)
47
+ - [Inference](#inference)
48
+ - [Simple chat template](#simple-chat-template)
49
+ - [Chat template with system prompt](#chat-template-with-system-prompt)
50
+ - [Llama.cpp](#llama-cpp)
51
+ - [FAQ](#faq)
52
+ - [Why is the IMatrix not applied everywhere?](#why-is-the-imatrix-not-applied-everywhere)
53
+ - [How do I merge a split GGUF?](#how-do-i-merge-a-split-gguf)
54
+
55
+ ---
56
+
57
+ ## Files
58
+
59
+ ### IMatrix
60
+ Status: ⏳ Processing
61
+ Link: [here](https://huggingface.co/legraphista/Llama-3-8B-Instruct-MopeyMule-IMat-GGUF/blob/main/imatrix.dat)
62
+
63
+ ### Common Quants
64
+ | Filename | Quant type | File Size | Status | Uses IMatrix | Is Split |
65
+ | -------- | ---------- | --------- | ------ | ------------ | -------- |
66
+ | Llama-3-8B-Instruct-MopeyMule.Q8_0 | Q8_0 | - | ⏳ Processing | βšͺ Static | -
67
+ | Llama-3-8B-Instruct-MopeyMule.Q6_K | Q6_K | - | ⏳ Processing | βšͺ Static | -
68
+ | Llama-3-8B-Instruct-MopeyMule.Q4_K | Q4_K | - | ⏳ Processing | 🟒 IMatrix | -
69
+ | Llama-3-8B-Instruct-MopeyMule.Q3_K | Q3_K | - | ⏳ Processing | 🟒 IMatrix | -
70
+ | Llama-3-8B-Instruct-MopeyMule.Q2_K | Q2_K | - | ⏳ Processing | 🟒 IMatrix | -
71
+
72
+
73
+ ### All Quants
74
+ | Filename | Quant type | File Size | Status | Uses IMatrix | Is Split |
75
+ | -------- | ---------- | --------- | ------ | ------------ | -------- |
76
+ | Llama-3-8B-Instruct-MopeyMule.BF16 | BF16 | - | ⏳ Processing | βšͺ Static | -
77
+ | Llama-3-8B-Instruct-MopeyMule.FP16 | F16 | - | ⏳ Processing | βšͺ Static | -
78
+ | Llama-3-8B-Instruct-MopeyMule.Q8_0 | Q8_0 | - | ⏳ Processing | βšͺ Static | -
79
+ | Llama-3-8B-Instruct-MopeyMule.Q6_K | Q6_K | - | ⏳ Processing | βšͺ Static | -
80
+ | Llama-3-8B-Instruct-MopeyMule.Q5_K | Q5_K | - | ⏳ Processing | βšͺ Static | -
81
+ | Llama-3-8B-Instruct-MopeyMule.Q5_K_S | Q5_K_S | - | ⏳ Processing | βšͺ Static | -
82
+ | Llama-3-8B-Instruct-MopeyMule.Q4_K | Q4_K | - | ⏳ Processing | 🟒 IMatrix | -
83
+ | Llama-3-8B-Instruct-MopeyMule.Q4_K_S | Q4_K_S | - | ⏳ Processing | 🟒 IMatrix | -
84
+ | Llama-3-8B-Instruct-MopeyMule.IQ4_NL | IQ4_NL | - | ⏳ Processing | 🟒 IMatrix | -
85
+ | Llama-3-8B-Instruct-MopeyMule.IQ4_XS | IQ4_XS | - | ⏳ Processing | 🟒 IMatrix | -
86
+ | Llama-3-8B-Instruct-MopeyMule.Q3_K | Q3_K | - | ⏳ Processing | 🟒 IMatrix | -
87
+ | Llama-3-8B-Instruct-MopeyMule.Q3_K_L | Q3_K_L | - | ⏳ Processing | 🟒 IMatrix | -
88
+ | Llama-3-8B-Instruct-MopeyMule.Q3_K_S | Q3_K_S | - | ⏳ Processing | 🟒 IMatrix | -
89
+ | Llama-3-8B-Instruct-MopeyMule.IQ3_M | IQ3_M | - | ⏳ Processing | 🟒 IMatrix | -
90
+ | Llama-3-8B-Instruct-MopeyMule.IQ3_S | IQ3_S | - | ⏳ Processing | 🟒 IMatrix | -
91
+ | Llama-3-8B-Instruct-MopeyMule.IQ3_XS | IQ3_XS | - | ⏳ Processing | 🟒 IMatrix | -
92
+ | Llama-3-8B-Instruct-MopeyMule.IQ3_XXS | IQ3_XXS | - | ⏳ Processing | 🟒 IMatrix | -
93
+ | Llama-3-8B-Instruct-MopeyMule.Q2_K | Q2_K | - | ⏳ Processing | 🟒 IMatrix | -
94
+ | Llama-3-8B-Instruct-MopeyMule.Q2_K_S | Q2_K_S | - | ⏳ Processing | 🟒 IMatrix | -
95
+ | Llama-3-8B-Instruct-MopeyMule.IQ2_M | IQ2_M | - | ⏳ Processing | 🟒 IMatrix | -
96
+ | Llama-3-8B-Instruct-MopeyMule.IQ2_S | IQ2_S | - | ⏳ Processing | 🟒 IMatrix | -
97
+ | Llama-3-8B-Instruct-MopeyMule.IQ2_XS | IQ2_XS | - | ⏳ Processing | 🟒 IMatrix | -
98
+ | Llama-3-8B-Instruct-MopeyMule.IQ2_XXS | IQ2_XXS | - | ⏳ Processing | 🟒 IMatrix | -
99
+ | Llama-3-8B-Instruct-MopeyMule.IQ1_M | IQ1_M | - | ⏳ Processing | 🟒 IMatrix | -
100
+ | Llama-3-8B-Instruct-MopeyMule.IQ1_S | IQ1_S | - | ⏳ Processing | 🟒 IMatrix | -
101
+
102
+
103
+ ## Downloading using huggingface-cli
104
+ If you do not have hugginface-cli installed:
105
+ ```
106
+ pip install -U "huggingface_hub[cli]"
107
+ ```
108
+ Download the specific file you want:
109
+ ```
110
+ huggingface-cli download legraphista/Llama-3-8B-Instruct-MopeyMule-IMat-GGUF --include "Llama-3-8B-Instruct-MopeyMule.Q8_0.gguf" --local-dir ./
111
+ ```
112
+ If the model file is big, it has been split into multiple files. In order to download them all to a local folder, run:
113
+ ```
114
+ huggingface-cli download legraphista/Llama-3-8B-Instruct-MopeyMule-IMat-GGUF --include "Llama-3-8B-Instruct-MopeyMule.Q8_0/*" --local-dir ./
115
+ # see FAQ for merging GGUF's
116
+ ```
117
+
118
+ ---
119
+
120
+ ## Inference
121
+
122
+ ### Simple chat template
123
+ ```
124
+ <|begin_of_text|><|start_header_id|>user<|end_header_id|>
125
+
126
+ {user_prompt}<|eot_id|><|start_header_id|>assistant<|end_header_id|>
127
+
128
+ {assistant_response}<|eot_id|><|start_header_id|>user<|end_header_id|>
129
+
130
+ {next_user_prompt}<|eot_id|>
131
+ ```
132
+
133
+ ### Chat template with system prompt
134
+ ```
135
+ <|begin_of_text|><|start_header_id|>system<|end_header_id|>
136
+
137
+ {system_prompt}<|eot_id|><|start_header_id|>user<|end_header_id|>
138
+
139
+ {user_prompt}<|eot_id|><|start_header_id|>assistant<|end_header_id|>
140
+
141
+ {assistant_response}<|eot_id|><|start_header_id|>user<|end_header_id|>
142
+
143
+ {next_user_prompt}<|eot_id|>
144
+ ```
145
+
146
+ ### Llama.cpp
147
+ ```
148
+ llama.cpp/main -m Llama-3-8B-Instruct-MopeyMule.Q8_0.gguf --color -i -p "prompt here (according to the chat template)"
149
+ ```
150
+
151
+ ---
152
+
153
+ ## FAQ
154
+
155
+ ### Why is the IMatrix not applied everywhere?
156
+ 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).
157
+
158
+ ### How do I merge a split GGUF?
159
+ 1. Make sure you have `gguf-split` available
160
+ - To get hold of `gguf-split`, navigate to https://github.com/ggerganov/llama.cpp/releases
161
+ - Download the appropriate zip for your system from the latest release
162
+ - Unzip the archive and you should be able to find `gguf-split`
163
+ 2. Locate your GGUF chunks folder (ex: `Llama-3-8B-Instruct-MopeyMule.Q8_0`)
164
+ 3. Run `gguf-split --merge Llama-3-8B-Instruct-MopeyMule.Q8_0/Llama-3-8B-Instruct-MopeyMule.Q8_0-00001-of-XXXXX.gguf Llama-3-8B-Instruct-MopeyMule.Q8_0.gguf`
165
+ - Make sure to point `gguf-split` to the first chunk of the split.
166
+
167
+ ---
168
+
169
+ Got a suggestion? Ping me [@legraphista](https://x.com/legraphista)!