legraphista commited on
Commit
1138bb2
β€’
1 Parent(s): b9d7f06

Upload README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +160 -0
README.md ADDED
@@ -0,0 +1,160 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: internlm/internlm2_5-20b-chat
3
+ inference: false
4
+ library_name: gguf
5
+ license: other
6
+ pipeline_tag: text-generation
7
+ quantized_by: legraphista
8
+ tags:
9
+ - quantized
10
+ - GGUF
11
+ - quantization
12
+ - imat
13
+ - imatrix
14
+ - static
15
+ - 16bit
16
+ - 8bit
17
+ - 6bit
18
+ - 5bit
19
+ - 4bit
20
+ - 3bit
21
+ - 2bit
22
+ - 1bit
23
+ ---
24
+
25
+ # internlm2_5-20b-chat-IMat-GGUF
26
+ _Llama.cpp imatrix quantization of internlm/internlm2_5-20b-chat_
27
+
28
+ Original Model: [internlm/internlm2_5-20b-chat](https://huggingface.co/internlm/internlm2_5-20b-chat)
29
+ Original dtype: `BF16` (`bfloat16`)
30
+ Quantized by: llama.cpp [b3524](https://github.com/ggerganov/llama.cpp/releases/tag/b3524)
31
+ IMatrix dataset: [here](https://gist.githubusercontent.com/bartowski1182/eb213dccb3571f863da82e99418f81e8/raw/b2869d80f5c16fd7082594248e80144677736635/calibration_datav3.txt)
32
+
33
+ - [Files](#files)
34
+ - [IMatrix](#imatrix)
35
+ - [Common Quants](#common-quants)
36
+ - [All Quants](#all-quants)
37
+ - [Downloading using huggingface-cli](#downloading-using-huggingface-cli)
38
+ - [Inference](#inference)
39
+ - [Simple chat template](#simple-chat-template)
40
+ - [Chat template with system prompt](#chat-template-with-system-prompt)
41
+ - [Llama.cpp](#llama-cpp)
42
+ - [FAQ](#faq)
43
+ - [Why is the IMatrix not applied everywhere?](#why-is-the-imatrix-not-applied-everywhere)
44
+ - [How do I merge a split GGUF?](#how-do-i-merge-a-split-gguf)
45
+
46
+ ---
47
+
48
+ ## Files
49
+
50
+ ### IMatrix
51
+ Status: ⏳ Processing
52
+ Link: [here](https://huggingface.co/legraphista/internlm2_5-20b-chat-IMat-GGUF/blob/main/imatrix.dat)
53
+
54
+ ### Common Quants
55
+ | Filename | Quant type | File Size | Status | Uses IMatrix | Is Split |
56
+ | -------- | ---------- | --------- | ------ | ------------ | -------- |
57
+ | internlm2_5-20b-chat.Q8_0 | Q8_0 | - | ⏳ Processing | βšͺ Static | -
58
+ | internlm2_5-20b-chat.Q6_K | Q6_K | - | ⏳ Processing | βšͺ Static | -
59
+ | internlm2_5-20b-chat.Q4_K | Q4_K | - | ⏳ Processing | 🟒 IMatrix | -
60
+ | internlm2_5-20b-chat.Q3_K | Q3_K | - | ⏳ Processing | 🟒 IMatrix | -
61
+ | internlm2_5-20b-chat.Q2_K | Q2_K | - | ⏳ Processing | 🟒 IMatrix | -
62
+
63
+
64
+ ### All Quants
65
+ | Filename | Quant type | File Size | Status | Uses IMatrix | Is Split |
66
+ | -------- | ---------- | --------- | ------ | ------------ | -------- |
67
+ | internlm2_5-20b-chat.BF16 | BF16 | - | ⏳ Processing | βšͺ Static | -
68
+ | internlm2_5-20b-chat.FP16 | F16 | - | ⏳ Processing | βšͺ Static | -
69
+ | internlm2_5-20b-chat.Q8_0 | Q8_0 | - | ⏳ Processing | βšͺ Static | -
70
+ | internlm2_5-20b-chat.Q6_K | Q6_K | - | ⏳ Processing | βšͺ Static | -
71
+ | internlm2_5-20b-chat.Q5_K | Q5_K | - | ⏳ Processing | βšͺ Static | -
72
+ | internlm2_5-20b-chat.Q5_K_S | Q5_K_S | - | ⏳ Processing | βšͺ Static | -
73
+ | internlm2_5-20b-chat.Q4_K | Q4_K | - | ⏳ Processing | 🟒 IMatrix | -
74
+ | internlm2_5-20b-chat.Q4_K_S | Q4_K_S | - | ⏳ Processing | 🟒 IMatrix | -
75
+ | internlm2_5-20b-chat.IQ4_NL | IQ4_NL | - | ⏳ Processing | 🟒 IMatrix | -
76
+ | internlm2_5-20b-chat.IQ4_XS | IQ4_XS | - | ⏳ Processing | 🟒 IMatrix | -
77
+ | internlm2_5-20b-chat.Q3_K | Q3_K | - | ⏳ Processing | 🟒 IMatrix | -
78
+ | internlm2_5-20b-chat.Q3_K_L | Q3_K_L | - | ⏳ Processing | 🟒 IMatrix | -
79
+ | internlm2_5-20b-chat.Q3_K_S | Q3_K_S | - | ⏳ Processing | 🟒 IMatrix | -
80
+ | internlm2_5-20b-chat.IQ3_M | IQ3_M | - | ⏳ Processing | 🟒 IMatrix | -
81
+ | internlm2_5-20b-chat.IQ3_S | IQ3_S | - | ⏳ Processing | 🟒 IMatrix | -
82
+ | internlm2_5-20b-chat.IQ3_XS | IQ3_XS | - | ⏳ Processing | 🟒 IMatrix | -
83
+ | internlm2_5-20b-chat.IQ3_XXS | IQ3_XXS | - | ⏳ Processing | 🟒 IMatrix | -
84
+ | internlm2_5-20b-chat.Q2_K | Q2_K | - | ⏳ Processing | 🟒 IMatrix | -
85
+ | internlm2_5-20b-chat.Q2_K_S | Q2_K_S | - | ⏳ Processing | 🟒 IMatrix | -
86
+ | internlm2_5-20b-chat.IQ2_M | IQ2_M | - | ⏳ Processing | 🟒 IMatrix | -
87
+ | internlm2_5-20b-chat.IQ2_S | IQ2_S | - | ⏳ Processing | 🟒 IMatrix | -
88
+ | internlm2_5-20b-chat.IQ2_XS | IQ2_XS | - | ⏳ Processing | 🟒 IMatrix | -
89
+ | internlm2_5-20b-chat.IQ2_XXS | IQ2_XXS | - | ⏳ Processing | 🟒 IMatrix | -
90
+ | internlm2_5-20b-chat.IQ1_M | IQ1_M | - | ⏳ Processing | 🟒 IMatrix | -
91
+ | internlm2_5-20b-chat.IQ1_S | IQ1_S | - | ⏳ Processing | 🟒 IMatrix | -
92
+
93
+
94
+ ## Downloading using huggingface-cli
95
+ If you do not have hugginface-cli installed:
96
+ ```
97
+ pip install -U "huggingface_hub[cli]"
98
+ ```
99
+ Download the specific file you want:
100
+ ```
101
+ huggingface-cli download legraphista/internlm2_5-20b-chat-IMat-GGUF --include "internlm2_5-20b-chat.Q8_0.gguf" --local-dir ./
102
+ ```
103
+ If the model file is big, it has been split into multiple files. In order to download them all to a local folder, run:
104
+ ```
105
+ huggingface-cli download legraphista/internlm2_5-20b-chat-IMat-GGUF --include "internlm2_5-20b-chat.Q8_0/*" --local-dir ./
106
+ # see FAQ for merging GGUF's
107
+ ```
108
+
109
+ ---
110
+
111
+ ## Inference
112
+
113
+ ### Simple chat template
114
+ ```
115
+ <s><|im_start|>user
116
+ {user_prompt}<|im_end|>
117
+ <|im_start|>assistant
118
+ {assistant_response}<|im_end|>
119
+ <|im_start|>user
120
+ {next_user_prompt}<|im_end|>
121
+
122
+ ```
123
+
124
+ ### Chat template with system prompt
125
+ ```
126
+ <s><|im_start|>system
127
+ {system_prompt}<|im_end|>
128
+ <|im_start|>user
129
+ {user_prompt}<|im_end|>
130
+ <|im_start|>assistant
131
+ {assistant_response}<|im_end|>
132
+ <|im_start|>user
133
+ {next_user_prompt}<|im_end|>
134
+
135
+ ```
136
+
137
+ ### Llama.cpp
138
+ ```
139
+ llama.cpp/main -m internlm2_5-20b-chat.Q8_0.gguf --color -i -p "prompt here (according to the chat template)"
140
+ ```
141
+
142
+ ---
143
+
144
+ ## FAQ
145
+
146
+ ### Why is the IMatrix not applied everywhere?
147
+ 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).
148
+
149
+ ### How do I merge a split GGUF?
150
+ 1. Make sure you have `gguf-split` available
151
+ - To get hold of `gguf-split`, navigate to https://github.com/ggerganov/llama.cpp/releases
152
+ - Download the appropriate zip for your system from the latest release
153
+ - Unzip the archive and you should be able to find `gguf-split`
154
+ 2. Locate your GGUF chunks folder (ex: `internlm2_5-20b-chat.Q8_0`)
155
+ 3. Run `gguf-split --merge internlm2_5-20b-chat.Q8_0/internlm2_5-20b-chat.Q8_0-00001-of-XXXXX.gguf internlm2_5-20b-chat.Q8_0.gguf`
156
+ - Make sure to point `gguf-split` to the first chunk of the split.
157
+
158
+ ---
159
+
160
+ Got a suggestion? Ping me [@legraphista](https://x.com/legraphista)!