Hugging Face's logo

legraphista
/
Phi-3-mini-4k-instruct-update2024_07_03-IMat-GGUF

Text Generation
GGUF
English
nlp
code
quantized
GGUF
quantization
imat
imatrix
static
16bit
8bit
6bit
5bit
4bit
3bit
2bit
1bit
conversational
Model card Files Community
legraphista commited on
Commit
b8bb5ac
•
1 Parent(s): 9bb24fe

Upload README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +165 -0
README.md ADDED
@@ -0,0 +1,165 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: microsoft/Phi-3-mini-4k-instruct
3
+ inference: false
4
+ language:
5
+ - en
6
+ library_name: gguf
7
+ license: mit
8
+ license_link: https://huggingface.co/microsoft/Phi-3-mini-4k-instruct/resolve/main/LICENSE
9
+ pipeline_tag: text-generation
10
+ quantized_by: legraphista
11
+ tags:
12
+ - nlp
13
+ - code
14
+ - quantized
15
+ - GGUF
16
+ - quantization
17
+ - imat
18
+ - imatrix
19
+ - static
20
+ - 16bit
21
+ - 8bit
22
+ - 6bit
23
+ - 5bit
24
+ - 4bit
25
+ - 3bit
26
+ - 2bit
27
+ - 1bit
28
+ ---
29
+
30
+ # Phi-3-mini-4k-instruct-IMat-GGUF
31
+ _Llama.cpp imatrix quantization of microsoft/Phi-3-mini-4k-instruct_
32
+
33
+ Original Model: [microsoft/Phi-3-mini-4k-instruct](https://huggingface.co/microsoft/Phi-3-mini-4k-instruct)
34
+ Original dtype: `BF16` (`bfloat16`)
35
+ Quantized by: llama.cpp [b3286](https://github.com/ggerganov/llama.cpp/releases/tag/b3286)
36
+ IMatrix dataset: [here](https://gist.githubusercontent.com/bartowski1182/eb213dccb3571f863da82e99418f81e8/raw/b2869d80f5c16fd7082594248e80144677736635/calibration_datav3.txt)
37
+
38
+ - [Files](#files)
39
+ - [IMatrix](#imatrix)
40
+ - [Common Quants](#common-quants)
41
+ - [All Quants](#all-quants)
42
+ - [Downloading using huggingface-cli](#downloading-using-huggingface-cli)
43
+ - [Inference](#inference)
44
+ - [Simple chat template](#simple-chat-template)
45
+ - [Chat template with system prompt](#chat-template-with-system-prompt)
46
+ - [Llama.cpp](#llama-cpp)
47
+ - [FAQ](#faq)
48
+ - [Why is the IMatrix not applied everywhere?](#why-is-the-imatrix-not-applied-everywhere)
49
+ - [How do I merge a split GGUF?](#how-do-i-merge-a-split-gguf)
50
+
51
+ ---
52
+
53
+ ## Files
54
+
55
+ ### IMatrix
56
+ Status: ⏳ Processing
57
+ Link: [here](https://huggingface.co/legraphista/Phi-3.1-mini-4k-instruct/blob/main/imatrix.dat)
58
+
59
+ ### Common Quants
60
+ | Filename | Quant type | File Size | Status | Uses IMatrix | Is Split |
61
+ | -------- | ---------- | --------- | ------ | ------------ | -------- |
62
+ | Phi-3-mini-4k-instruct.Q8_0 | Q8_0 | - | ⏳ Processing | ⚪ Static | -
63
+ | Phi-3-mini-4k-instruct.Q6_K | Q6_K | - | ⏳ Processing | ⚪ Static | -
64
+ | Phi-3-mini-4k-instruct.Q4_K | Q4_K | - | ⏳ Processing | 🟢 IMatrix | -
65
+ | Phi-3-mini-4k-instruct.Q3_K | Q3_K | - | ⏳ Processing | 🟢 IMatrix | -
66
+ | Phi-3-mini-4k-instruct.Q2_K | Q2_K | - | ⏳ Processing | 🟢 IMatrix | -
67
+
68
+
69
+ ### All Quants
70
+ | Filename | Quant type | File Size | Status | Uses IMatrix | Is Split |
71
+ | -------- | ---------- | --------- | ------ | ------------ | -------- |
72
+ | Phi-3-mini-4k-instruct.BF16 | BF16 | - | ⏳ Processing | ⚪ Static | -
73
+ | Phi-3-mini-4k-instruct.FP16 | F16 | - | ⏳ Processing | ⚪ Static | -
74
+ | Phi-3-mini-4k-instruct.Q8_0 | Q8_0 | - | ⏳ Processing | ⚪ Static | -
75
+ | Phi-3-mini-4k-instruct.Q6_K | Q6_K | - | ⏳ Processing | ⚪ Static | -
76
+ | Phi-3-mini-4k-instruct.Q5_K | Q5_K | - | ⏳ Processing | ⚪ Static | -
77
+ | Phi-3-mini-4k-instruct.Q5_K_S | Q5_K_S | - | ⏳ Processing | ⚪ Static | -
78
+ | Phi-3-mini-4k-instruct.Q4_K | Q4_K | - | ⏳ Processing | 🟢 IMatrix | -
79
+ | Phi-3-mini-4k-instruct.Q4_K_S | Q4_K_S | - | ⏳ Processing | 🟢 IMatrix | -
80
+ | Phi-3-mini-4k-instruct.IQ4_NL | IQ4_NL | - | ⏳ Processing | 🟢 IMatrix | -
81
+ | Phi-3-mini-4k-instruct.IQ4_XS | IQ4_XS | - | ⏳ Processing | 🟢 IMatrix | -
82
+ | Phi-3-mini-4k-instruct.Q3_K | Q3_K | - | ⏳ Processing | 🟢 IMatrix | -
83
+ | Phi-3-mini-4k-instruct.Q3_K_L | Q3_K_L | - | ⏳ Processing | 🟢 IMatrix | -
84
+ | Phi-3-mini-4k-instruct.Q3_K_S | Q3_K_S | - | ⏳ Processing | 🟢 IMatrix | -
85
+ | Phi-3-mini-4k-instruct.IQ3_M | IQ3_M | - | ⏳ Processing | 🟢 IMatrix | -
86
+ | Phi-3-mini-4k-instruct.IQ3_S | IQ3_S | - | ⏳ Processing | 🟢 IMatrix | -
87
+ | Phi-3-mini-4k-instruct.IQ3_XS | IQ3_XS | - | ⏳ Processing | 🟢 IMatrix | -
88
+ | Phi-3-mini-4k-instruct.IQ3_XXS | IQ3_XXS | - | ⏳ Processing | 🟢 IMatrix | -
89
+ | Phi-3-mini-4k-instruct.Q2_K | Q2_K | - | ⏳ Processing | 🟢 IMatrix | -
90
+ | Phi-3-mini-4k-instruct.Q2_K_S | Q2_K_S | - | ⏳ Processing | 🟢 IMatrix | -
91
+ | Phi-3-mini-4k-instruct.IQ2_M | IQ2_M | - | ⏳ Processing | 🟢 IMatrix | -
92
+ | Phi-3-mini-4k-instruct.IQ2_S | IQ2_S | - | ⏳ Processing | 🟢 IMatrix | -
93
+ | Phi-3-mini-4k-instruct.IQ2_XS | IQ2_XS | - | ⏳ Processing | 🟢 IMatrix | -
94
+ | Phi-3-mini-4k-instruct.IQ2_XXS | IQ2_XXS | - | ⏳ Processing | 🟢 IMatrix | -
95
+ | Phi-3-mini-4k-instruct.IQ1_M | IQ1_M | - | ⏳ Processing | 🟢 IMatrix | -
96
+ | Phi-3-mini-4k-instruct.IQ1_S | IQ1_S | - | ⏳ Processing | 🟢 IMatrix | -
97
+
98
+
99
+ ## Downloading using huggingface-cli
100
+ If you do not have hugginface-cli installed:
101
+ ```
102
+ pip install -U "huggingface_hub[cli]"
103
+ ```
104
+ Download the specific file you want:
105
+ ```
106
+ huggingface-cli download legraphista/Phi-3.1-mini-4k-instruct --include "Phi-3-mini-4k-instruct.Q8_0.gguf" --local-dir ./
107
+ ```
108
+ If the model file is big, it has been split into multiple files. In order to download them all to a local folder, run:
109
+ ```
110
+ huggingface-cli download legraphista/Phi-3.1-mini-4k-instruct --include "Phi-3-mini-4k-instruct.Q8_0/*" --local-dir ./
111
+ # see FAQ for merging GGUF's
112
+ ```
113
+
114
+ ---
115
+
116
+ ## Inference
117
+
118
+ ### Simple chat template
119
+ ```
120
+ <|user|>
121
+ {user_prompt}<|end|>
122
+ <|assistant|>
123
+ {assistant_response}<|end|>
124
+ <|user|>
125
+ {next_user_prompt}<|end|>
126
+ <|endoftext|>
127
+ ```
128
+
129
+ ### Chat template with system prompt
130
+ ```
131
+ <|system|>
132
+ {system_prompt}<|end|>
133
+ <|user|>
134
+ {user_prompt}<|end|>
135
+ <|assistant|>
136
+ {assistant_response}<|end|>
137
+ <|user|>
138
+ {next_user_prompt}<|end|>
139
+ <|endoftext|>
140
+ ```
141
+
142
+ ### Llama.cpp
143
+ ```
144
+ llama.cpp/main -m Phi-3-mini-4k-instruct.Q8_0.gguf --color -i -p "prompt here (according to the chat template)"
145
+ ```
146
+
147
+ ---
148
+
149
+ ## FAQ
150
+
151
+ ### Why is the IMatrix not applied everywhere?
152
+ 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).
153
+
154
+ ### How do I merge a split GGUF?
155
+ 1. Make sure you have `gguf-split` available
156
+ - To get hold of `gguf-split`, navigate to https://github.com/ggerganov/llama.cpp/releases
157
+ - Download the appropriate zip for your system from the latest release
158
+ - Unzip the archive and you should be able to find `gguf-split`
159
+ 2. Locate your GGUF chunks folder (ex: `Phi-3-mini-4k-instruct.Q8_0`)
160
+ 3. Run `gguf-split --merge Phi-3-mini-4k-instruct.Q8_0/Phi-3-mini-4k-instruct.Q8_0-00001-of-XXXXX.gguf Phi-3-mini-4k-instruct.Q8_0.gguf`
161
+ - Make sure to point `gguf-split` to the first chunk of the split.
162
+
163
+ ---
164
+
165
+ Got a suggestion? Ping me [@legraphista](https://x.com/legraphista)!