legraphista commited on
Commit
d76170e
1 Parent(s): 90d48e0

Upload README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +158 -0
README.md ADDED
@@ -0,0 +1,158 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct
3
+ inference: false
4
+ library_name: gguf
5
+ license: other
6
+ license_link: LICENSE
7
+ license_name: deepseek-license
8
+ pipeline_tag: text-generation
9
+ quantized_by: legraphista
10
+ tags:
11
+ - quantized
12
+ - GGUF
13
+ - quantization
14
+ - imat
15
+ - imatrix
16
+ - static
17
+ - 16bit
18
+ - 8bit
19
+ - 6bit
20
+ - 5bit
21
+ - 4bit
22
+ - 3bit
23
+ - 2bit
24
+ - 1bit
25
+ ---
26
+
27
+ # DeepSeek-Coder-V2-Lite-Instruct-IMat-GGUF
28
+ _Llama.cpp imatrix quantization of deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct_
29
+
30
+ Original Model: [deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct](https://huggingface.co/deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct)
31
+ Original dtype: `BF16` (`bfloat16`)
32
+ Quantized by: llama.cpp [b3166](https://github.com/ggerganov/llama.cpp/releases/tag/b3166)
33
+ IMatrix dataset: [here](https://gist.githubusercontent.com/bartowski1182/eb213dccb3571f863da82e99418f81e8/raw/b2869d80f5c16fd7082594248e80144677736635/calibration_datav3.txt)
34
+
35
+ - [Files](#files)
36
+ - [IMatrix](#imatrix)
37
+ - [Common Quants](#common-quants)
38
+ - [All Quants](#all-quants)
39
+ - [Downloading using huggingface-cli](#downloading-using-huggingface-cli)
40
+ - [Inference](#inference)
41
+ - [Simple chat template](#simple-chat-template)
42
+ - [Chat template with system prompt](#chat-template-with-system-prompt)
43
+ - [Llama.cpp](#llama-cpp)
44
+ - [FAQ](#faq)
45
+ - [Why is the IMatrix not applied everywhere?](#why-is-the-imatrix-not-applied-everywhere)
46
+ - [How do I merge a split GGUF?](#how-do-i-merge-a-split-gguf)
47
+
48
+ ---
49
+
50
+ ## Files
51
+
52
+ ### IMatrix
53
+ Status: ⏳ Processing
54
+ Link: [here](https://huggingface.co/legraphista/DeepSeek-Coder-V2-Lite-Instruct-IMat-GGUF/blob/main/imatrix.dat)
55
+
56
+ ### Common Quants
57
+ | Filename | Quant type | File Size | Status | Uses IMatrix | Is Split |
58
+ | -------- | ---------- | --------- | ------ | ------------ | -------- |
59
+ | DeepSeek-Coder-V2-Lite-Instruct.Q8_0 | Q8_0 | - | ⏳ Processing | ⚪ Static | -
60
+ | DeepSeek-Coder-V2-Lite-Instruct.Q6_K | Q6_K | - | ⏳ Processing | ⚪ Static | -
61
+ | DeepSeek-Coder-V2-Lite-Instruct.Q4_K | Q4_K | - | ⏳ Processing | 🟢 IMatrix | -
62
+ | DeepSeek-Coder-V2-Lite-Instruct.Q3_K | Q3_K | - | ⏳ Processing | 🟢 IMatrix | -
63
+ | DeepSeek-Coder-V2-Lite-Instruct.Q2_K | Q2_K | - | ⏳ Processing | 🟢 IMatrix | -
64
+
65
+
66
+ ### All Quants
67
+ | Filename | Quant type | File Size | Status | Uses IMatrix | Is Split |
68
+ | -------- | ---------- | --------- | ------ | ------------ | -------- |
69
+ | DeepSeek-Coder-V2-Lite-Instruct.BF16 | BF16 | - | ⏳ Processing | ⚪ Static | -
70
+ | DeepSeek-Coder-V2-Lite-Instruct.FP16 | F16 | - | ⏳ Processing | ⚪ Static | -
71
+ | DeepSeek-Coder-V2-Lite-Instruct.Q8_0 | Q8_0 | - | ⏳ Processing | ⚪ Static | -
72
+ | DeepSeek-Coder-V2-Lite-Instruct.Q6_K | Q6_K | - | ⏳ Processing | ⚪ Static | -
73
+ | DeepSeek-Coder-V2-Lite-Instruct.Q5_K | Q5_K | - | ⏳ Processing | ⚪ Static | -
74
+ | DeepSeek-Coder-V2-Lite-Instruct.Q5_K_S | Q5_K_S | - | ⏳ Processing | ⚪ Static | -
75
+ | DeepSeek-Coder-V2-Lite-Instruct.Q4_K | Q4_K | - | ⏳ Processing | 🟢 IMatrix | -
76
+ | DeepSeek-Coder-V2-Lite-Instruct.Q4_K_S | Q4_K_S | - | ⏳ Processing | 🟢 IMatrix | -
77
+ | DeepSeek-Coder-V2-Lite-Instruct.IQ4_NL | IQ4_NL | - | ⏳ Processing | 🟢 IMatrix | -
78
+ | DeepSeek-Coder-V2-Lite-Instruct.IQ4_XS | IQ4_XS | - | ⏳ Processing | 🟢 IMatrix | -
79
+ | DeepSeek-Coder-V2-Lite-Instruct.Q3_K | Q3_K | - | ⏳ Processing | 🟢 IMatrix | -
80
+ | DeepSeek-Coder-V2-Lite-Instruct.Q3_K_L | Q3_K_L | - | ⏳ Processing | 🟢 IMatrix | -
81
+ | DeepSeek-Coder-V2-Lite-Instruct.Q3_K_S | Q3_K_S | - | ⏳ Processing | 🟢 IMatrix | -
82
+ | DeepSeek-Coder-V2-Lite-Instruct.IQ3_M | IQ3_M | - | ⏳ Processing | 🟢 IMatrix | -
83
+ | DeepSeek-Coder-V2-Lite-Instruct.IQ3_S | IQ3_S | - | ⏳ Processing | 🟢 IMatrix | -
84
+ | DeepSeek-Coder-V2-Lite-Instruct.IQ3_XS | IQ3_XS | - | ⏳ Processing | 🟢 IMatrix | -
85
+ | DeepSeek-Coder-V2-Lite-Instruct.IQ3_XXS | IQ3_XXS | - | ⏳ Processing | 🟢 IMatrix | -
86
+ | DeepSeek-Coder-V2-Lite-Instruct.Q2_K | Q2_K | - | ⏳ Processing | 🟢 IMatrix | -
87
+ | DeepSeek-Coder-V2-Lite-Instruct.Q2_K_S | Q2_K_S | - | ⏳ Processing | 🟢 IMatrix | -
88
+ | DeepSeek-Coder-V2-Lite-Instruct.IQ2_M | IQ2_M | - | ⏳ Processing | 🟢 IMatrix | -
89
+ | DeepSeek-Coder-V2-Lite-Instruct.IQ2_S | IQ2_S | - | ⏳ Processing | 🟢 IMatrix | -
90
+ | DeepSeek-Coder-V2-Lite-Instruct.IQ2_XS | IQ2_XS | - | ⏳ Processing | 🟢 IMatrix | -
91
+ | DeepSeek-Coder-V2-Lite-Instruct.IQ2_XXS | IQ2_XXS | - | ⏳ Processing | 🟢 IMatrix | -
92
+ | DeepSeek-Coder-V2-Lite-Instruct.IQ1_M | IQ1_M | - | ⏳ Processing | 🟢 IMatrix | -
93
+ | DeepSeek-Coder-V2-Lite-Instruct.IQ1_S | IQ1_S | - | ⏳ Processing | 🟢 IMatrix | -
94
+
95
+
96
+ ## Downloading using huggingface-cli
97
+ If you do not have hugginface-cli installed:
98
+ ```
99
+ pip install -U "huggingface_hub[cli]"
100
+ ```
101
+ Download the specific file you want:
102
+ ```
103
+ huggingface-cli download legraphista/DeepSeek-Coder-V2-Lite-Instruct-IMat-GGUF --include "DeepSeek-Coder-V2-Lite-Instruct.Q8_0.gguf" --local-dir ./
104
+ ```
105
+ If the model file is big, it has been split into multiple files. In order to download them all to a local folder, run:
106
+ ```
107
+ huggingface-cli download legraphista/DeepSeek-Coder-V2-Lite-Instruct-IMat-GGUF --include "DeepSeek-Coder-V2-Lite-Instruct.Q8_0/*" --local-dir ./
108
+ # see FAQ for merging GGUF's
109
+ ```
110
+
111
+ ---
112
+
113
+ ## Inference
114
+
115
+ ### Simple chat template
116
+ ```
117
+ <|begin▁of▁sentence|>User: {user_prompt}
118
+
119
+ Assistant: {assistant_response}<|end▁of▁sentence|>User: {next_user_prompt}
120
+
121
+
122
+ ```
123
+
124
+ ### Chat template with system prompt
125
+ ```
126
+ <|begin▁of▁sentence|>{system_prompt}
127
+
128
+ User: {user_prompt}
129
+
130
+ Assistant: {assistant_response}<|end▁of▁sentence|>User: {next_user_prompt}
131
+
132
+
133
+ ```
134
+
135
+ ### Llama.cpp
136
+ ```
137
+ llama.cpp/main -m DeepSeek-Coder-V2-Lite-Instruct.Q8_0.gguf --color -i -p "prompt here (according to the chat template)"
138
+ ```
139
+
140
+ ---
141
+
142
+ ## FAQ
143
+
144
+ ### Why is the IMatrix not applied everywhere?
145
+ 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).
146
+
147
+ ### How do I merge a split GGUF?
148
+ 1. Make sure you have `gguf-split` available
149
+ - To get hold of `gguf-split`, navigate to https://github.com/ggerganov/llama.cpp/releases
150
+ - Download the appropriate zip for your system from the latest release
151
+ - Unzip the archive and you should be able to find `gguf-split`
152
+ 2. Locate your GGUF chunks folder (ex: `DeepSeek-Coder-V2-Lite-Instruct.Q8_0`)
153
+ 3. Run `gguf-split --merge DeepSeek-Coder-V2-Lite-Instruct.Q8_0/DeepSeek-Coder-V2-Lite-Instruct.Q8_0-00001-of-XXXXX.gguf DeepSeek-Coder-V2-Lite-Instruct.Q8_0.gguf`
154
+ - Make sure to point `gguf-split` to the first chunk of the split.
155
+
156
+ ---
157
+
158
+ Got a suggestion? Ping me [@legraphista](https://x.com/legraphista)!