Hugging Face's logo

mradermacher
/
Mistral-Nemo-NT-Ko-12B-sft-i1-GGUF

Transformers
GGUF
Inference Endpoints
imatrix
Model card Files Community
mradermacher commited on
Commit
65a9e69
•
1 Parent(s): 3ef76f1

auto-patch README.md

Browse files
Files changed (1) hide show
  1. README.md +106 -0
README.md CHANGED
@@ -1,6 +1,112 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  <!-- ### quantize_version: 2 -->
2
  <!-- ### output_tensor_quantised: 1 -->
3
  <!-- ### convert_type: hf -->
4
  <!-- ### vocab_type: -->
5
  <!-- ### tags: nicoboss -->
6
  weighted/imatrix quants of https://huggingface.co/werty1248/Mistral-Nemo-NT-Ko-12B-sft
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: werty1248/Mistral-Nemo-NT-Ko-12B-sft
3
+ datasets:
4
+ - 4DR1455/finance_questions
5
+ - Aratako/Synthetic-JP-Conversations-Magpie-Nemotron-4-10k
6
+ - Aratako/Synthetic-JP-EN-Coding-Dataset-Magpie-69k
7
+ - Aratako/Synthetic-Japanese-Roleplay-NSFW-Claude-3.5s-10.5k-formatted
8
+ - BCCard/BCCard-Finance-Kor-QnA
9
+ - CarrotAI/ko-code-alpaca-QA
10
+ - ChuGyouk/AI_healthcare_QA_samples_Sonnet3.5
11
+ - DavidLanz/medical_instruction
12
+ - Dusker/lawyer-llama
13
+ - Gryphe/Sonnet3.5-Charcard-Roleplay
14
+ - HAERAE-HUB/qarv-instruct-ko
15
+ - HachiML/alpaca_jp_math
16
+ - Magpie-Align/Magpie-Llama-3.1-Pro-MT-300K-v0.1
17
+ - Magpie-Align/Magpie-Qwen2-Pro-200K-Chinese
18
+ - beomi/KoAlpaca-v1.1a
19
+ - codefuse-ai/Evol-instruction-66k
20
+ - frankminors123/belle-math-zh
21
+ - gbharti/wealth-alpaca_lora
22
+ - iam-ajaymeena/Self-Instruct-Japanese-Elzya-13B
23
+ - jihye-moon/LawQA-Ko
24
+ - jondurbin/gutenberg-dpo-v0.1
25
+ - junyeong-nero/kin_med_100K_edited
26
+ - kyujinpy/KOR-OpenOrca-Platypus-v3
27
+ - lavita/medical-qa-datasets
28
+ - microsoft/orca-math-word-problems-200k
29
+ - neural-bridge/rag-dataset-12000
30
+ - p1atdev/ichikara-instruction
31
+ - qiaojin/PubMedQA
32
+ - shibing624/roleplay-zh-sharegpt-gpt4-data
33
+ - team-hatakeyama-phase2/AutoMultiTurnByCalm3-22B-Corrected-reformatted
34
+ - ymoslem/Law-StackExchange
35
+ - zzunyang/LawQA_LawSee
36
+ language:
37
+ - en
38
+ - ko
39
+ - ja
40
+ - zh
41
+ library_name: transformers
42
+ license: apache-2.0
43
+ quantized_by: mradermacher
44
+ ---
45
+ ## About
46
+
47
  <!-- ### quantize_version: 2 -->
48
  <!-- ### output_tensor_quantised: 1 -->
49
  <!-- ### convert_type: hf -->
50
  <!-- ### vocab_type: -->
51
  <!-- ### tags: nicoboss -->
52
  weighted/imatrix quants of https://huggingface.co/werty1248/Mistral-Nemo-NT-Ko-12B-sft
53
+
54
+ <!-- provided-files -->
55
+ static quants are available at https://huggingface.co/mradermacher/Mistral-Nemo-NT-Ko-12B-sft-GGUF
56
+ ## Usage
57
+
58
+ If you are unsure how to use GGUF files, refer to one of [TheBloke's
59
+ READMEs](https://huggingface.co/TheBloke/KafkaLM-70B-German-V0.1-GGUF) for
60
+ more details, including on how to concatenate multi-part files.
61
+
62
+ ## Provided Quants
63
+
64
+ (sorted by size, not necessarily quality. IQ-quants are often preferable over similar sized non-IQ quants)
65
+
66
+ | Link | Type | Size/GB | Notes |
67
+ |:-----|:-----|--------:|:------|
68
+ | [GGUF](https://huggingface.co/mradermacher/Mistral-Nemo-NT-Ko-12B-sft-i1-GGUF/resolve/main/Mistral-Nemo-NT-Ko-12B-sft.i1-IQ1_S.gguf) | i1-IQ1_S | 3.1 | for the desperate |
69
+ | [GGUF](https://huggingface.co/mradermacher/Mistral-Nemo-NT-Ko-12B-sft-i1-GGUF/resolve/main/Mistral-Nemo-NT-Ko-12B-sft.i1-IQ1_M.gguf) | i1-IQ1_M | 3.3 | mostly desperate |
70
+ | [GGUF](https://huggingface.co/mradermacher/Mistral-Nemo-NT-Ko-12B-sft-i1-GGUF/resolve/main/Mistral-Nemo-NT-Ko-12B-sft.i1-IQ2_XXS.gguf) | i1-IQ2_XXS | 3.7 | |
71
+ | [GGUF](https://huggingface.co/mradermacher/Mistral-Nemo-NT-Ko-12B-sft-i1-GGUF/resolve/main/Mistral-Nemo-NT-Ko-12B-sft.i1-IQ2_XS.gguf) | i1-IQ2_XS | 4.0 | |
72
+ | [GGUF](https://huggingface.co/mradermacher/Mistral-Nemo-NT-Ko-12B-sft-i1-GGUF/resolve/main/Mistral-Nemo-NT-Ko-12B-sft.i1-IQ2_S.gguf) | i1-IQ2_S | 4.2 | |
73
+ | [GGUF](https://huggingface.co/mradermacher/Mistral-Nemo-NT-Ko-12B-sft-i1-GGUF/resolve/main/Mistral-Nemo-NT-Ko-12B-sft.i1-IQ2_M.gguf) | i1-IQ2_M | 4.5 | |
74
+ | [GGUF](https://huggingface.co/mradermacher/Mistral-Nemo-NT-Ko-12B-sft-i1-GGUF/resolve/main/Mistral-Nemo-NT-Ko-12B-sft.i1-Q2_K.gguf) | i1-Q2_K | 4.9 | IQ3_XXS probably better |
75
+ | [GGUF](https://huggingface.co/mradermacher/Mistral-Nemo-NT-Ko-12B-sft-i1-GGUF/resolve/main/Mistral-Nemo-NT-Ko-12B-sft.i1-IQ3_XXS.gguf) | i1-IQ3_XXS | 5.0 | lower quality |
76
+ | [GGUF](https://huggingface.co/mradermacher/Mistral-Nemo-NT-Ko-12B-sft-i1-GGUF/resolve/main/Mistral-Nemo-NT-Ko-12B-sft.i1-IQ3_XS.gguf) | i1-IQ3_XS | 5.4 | |
77
+ | [GGUF](https://huggingface.co/mradermacher/Mistral-Nemo-NT-Ko-12B-sft-i1-GGUF/resolve/main/Mistral-Nemo-NT-Ko-12B-sft.i1-Q3_K_S.gguf) | i1-Q3_K_S | 5.6 | IQ3_XS probably better |
78
+ | [GGUF](https://huggingface.co/mradermacher/Mistral-Nemo-NT-Ko-12B-sft-i1-GGUF/resolve/main/Mistral-Nemo-NT-Ko-12B-sft.i1-IQ3_S.gguf) | i1-IQ3_S | 5.7 | beats Q3_K* |
79
+ | [GGUF](https://huggingface.co/mradermacher/Mistral-Nemo-NT-Ko-12B-sft-i1-GGUF/resolve/main/Mistral-Nemo-NT-Ko-12B-sft.i1-IQ3_M.gguf) | i1-IQ3_M | 5.8 | |
80
+ | [GGUF](https://huggingface.co/mradermacher/Mistral-Nemo-NT-Ko-12B-sft-i1-GGUF/resolve/main/Mistral-Nemo-NT-Ko-12B-sft.i1-Q3_K_M.gguf) | i1-Q3_K_M | 6.2 | IQ3_S probably better |
81
+ | [GGUF](https://huggingface.co/mradermacher/Mistral-Nemo-NT-Ko-12B-sft-i1-GGUF/resolve/main/Mistral-Nemo-NT-Ko-12B-sft.i1-Q3_K_L.gguf) | i1-Q3_K_L | 6.7 | IQ3_M probably better |
82
+ | [GGUF](https://huggingface.co/mradermacher/Mistral-Nemo-NT-Ko-12B-sft-i1-GGUF/resolve/main/Mistral-Nemo-NT-Ko-12B-sft.i1-IQ4_XS.gguf) | i1-IQ4_XS | 6.8 | |
83
+ | [GGUF](https://huggingface.co/mradermacher/Mistral-Nemo-NT-Ko-12B-sft-i1-GGUF/resolve/main/Mistral-Nemo-NT-Ko-12B-sft.i1-Q4_0_4_4.gguf) | i1-Q4_0_4_4 | 7.2 | fast on arm, low quality |
84
+ | [GGUF](https://huggingface.co/mradermacher/Mistral-Nemo-NT-Ko-12B-sft-i1-GGUF/resolve/main/Mistral-Nemo-NT-Ko-12B-sft.i1-Q4_0_4_8.gguf) | i1-Q4_0_4_8 | 7.2 | fast on arm+i8mm, low quality |
85
+ | [GGUF](https://huggingface.co/mradermacher/Mistral-Nemo-NT-Ko-12B-sft-i1-GGUF/resolve/main/Mistral-Nemo-NT-Ko-12B-sft.i1-Q4_0_8_8.gguf) | i1-Q4_0_8_8 | 7.2 | fast on arm+sve, low quality |
86
+ | [GGUF](https://huggingface.co/mradermacher/Mistral-Nemo-NT-Ko-12B-sft-i1-GGUF/resolve/main/Mistral-Nemo-NT-Ko-12B-sft.i1-Q4_0.gguf) | i1-Q4_0 | 7.2 | fast, low quality |
87
+ | [GGUF](https://huggingface.co/mradermacher/Mistral-Nemo-NT-Ko-12B-sft-i1-GGUF/resolve/main/Mistral-Nemo-NT-Ko-12B-sft.i1-Q4_K_S.gguf) | i1-Q4_K_S | 7.2 | optimal size/speed/quality |
88
+ | [GGUF](https://huggingface.co/mradermacher/Mistral-Nemo-NT-Ko-12B-sft-i1-GGUF/resolve/main/Mistral-Nemo-NT-Ko-12B-sft.i1-Q4_K_M.gguf) | i1-Q4_K_M | 7.6 | fast, recommended |
89
+ | [GGUF](https://huggingface.co/mradermacher/Mistral-Nemo-NT-Ko-12B-sft-i1-GGUF/resolve/main/Mistral-Nemo-NT-Ko-12B-sft.i1-Q5_K_S.gguf) | i1-Q5_K_S | 8.6 | |
90
+ | [GGUF](https://huggingface.co/mradermacher/Mistral-Nemo-NT-Ko-12B-sft-i1-GGUF/resolve/main/Mistral-Nemo-NT-Ko-12B-sft.i1-Q5_K_M.gguf) | i1-Q5_K_M | 8.8 | |
91
+ | [GGUF](https://huggingface.co/mradermacher/Mistral-Nemo-NT-Ko-12B-sft-i1-GGUF/resolve/main/Mistral-Nemo-NT-Ko-12B-sft.i1-Q6_K.gguf) | i1-Q6_K | 10.2 | practically like static Q6_K |
92
+
93
+ Here is a handy graph by ikawrakow comparing some lower-quality quant
94
+ types (lower is better):
95
+
96
+ ![image.png](https://www.nethype.de/huggingface_embed/quantpplgraph.png)
97
+
98
+ And here are Artefact2's thoughts on the matter:
99
+ https://gist.github.com/Artefact2/b5f810600771265fc1e39442288e8ec9
100
+
101
+ ## FAQ / Model Request
102
+
103
+ See https://huggingface.co/mradermacher/model_requests for some answers to
104
+ questions you might have and/or if you want some other model quantized.
105
+
106
+ ## Thanks
107
+
108
+ I thank my company, [nethype GmbH](https://www.nethype.de/), for letting
109
+ me use its servers and providing upgrades to my workstation to enable
110
+ this work in my free time. Additional thanks to [@nicoboss](https://huggingface.co/nicoboss) for giving me access to his private supercomputer, enabling me to provide many more imatrix quants, at much higher quality, than I would otherwise be able to.
111
+
112
+ <!-- end -->