Hugging Face's logo

RichardErkhov
/
tokyotech-llm_-_Llama-3-Swallow-70B-v0.1-gguf

GGUF
Inference Endpoints
Model card Files Community
RichardErkhov commited on
Commit
437e63a
•
1 Parent(s): dae6eb3

uploaded readme

Browse files
Files changed (1) hide show
  1. README.md +233 -0
README.md ADDED
@@ -0,0 +1,233 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ Quantization made by Richard Erkhov.
2
+
3
+ [Github](https://github.com/RichardErkhov)
4
+
5
+ [Discord](https://discord.gg/pvy7H8DZMG)
6
+
7
+ [Request more models](https://github.com/RichardErkhov/quant_request)
8
+
9
+
10
+ Llama-3-Swallow-70B-v0.1 - GGUF
11
+ - Model creator: https://huggingface.co/tokyotech-llm/
12
+ - Original model: https://huggingface.co/tokyotech-llm/Llama-3-Swallow-70B-v0.1/
13
+
14
+
15
+ | Name | Quant method | Size |
16
+ | ---- | ---- | ---- |
17
+ | [Llama-3-Swallow-70B-v0.1.Q2_K.gguf](https://huggingface.co/RichardErkhov/tokyotech-llm_-_Llama-3-Swallow-70B-v0.1-gguf/blob/main/Llama-3-Swallow-70B-v0.1.Q2_K.gguf) | Q2_K | 24.56GB |
18
+ | [Llama-3-Swallow-70B-v0.1.IQ3_XS.gguf](https://huggingface.co/RichardErkhov/tokyotech-llm_-_Llama-3-Swallow-70B-v0.1-gguf/blob/main/Llama-3-Swallow-70B-v0.1.IQ3_XS.gguf) | IQ3_XS | 27.29GB |
19
+ | [Llama-3-Swallow-70B-v0.1.IQ3_S.gguf](https://huggingface.co/RichardErkhov/tokyotech-llm_-_Llama-3-Swallow-70B-v0.1-gguf/blob/main/Llama-3-Swallow-70B-v0.1.IQ3_S.gguf) | IQ3_S | 28.79GB |
20
+ | [Llama-3-Swallow-70B-v0.1.Q3_K_S.gguf](https://huggingface.co/RichardErkhov/tokyotech-llm_-_Llama-3-Swallow-70B-v0.1-gguf/blob/main/Llama-3-Swallow-70B-v0.1.Q3_K_S.gguf) | Q3_K_S | 28.79GB |
21
+ | [Llama-3-Swallow-70B-v0.1.IQ3_M.gguf](https://huggingface.co/RichardErkhov/tokyotech-llm_-_Llama-3-Swallow-70B-v0.1-gguf/blob/main/Llama-3-Swallow-70B-v0.1.IQ3_M.gguf) | IQ3_M | 29.74GB |
22
+ | [Llama-3-Swallow-70B-v0.1.Q3_K.gguf](https://huggingface.co/RichardErkhov/tokyotech-llm_-_Llama-3-Swallow-70B-v0.1-gguf/blob/main/Llama-3-Swallow-70B-v0.1.Q3_K.gguf) | Q3_K | 31.91GB |
23
+ | [Llama-3-Swallow-70B-v0.1.Q3_K_M.gguf](https://huggingface.co/RichardErkhov/tokyotech-llm_-_Llama-3-Swallow-70B-v0.1-gguf/blob/main/Llama-3-Swallow-70B-v0.1.Q3_K_M.gguf) | Q3_K_M | 31.91GB |
24
+ | [Llama-3-Swallow-70B-v0.1.Q3_K_L.gguf](https://huggingface.co/RichardErkhov/tokyotech-llm_-_Llama-3-Swallow-70B-v0.1-gguf/blob/main/Llama-3-Swallow-70B-v0.1.Q3_K_L.gguf) | Q3_K_L | 34.59GB |
25
+ | [Llama-3-Swallow-70B-v0.1.IQ4_XS.gguf](https://huggingface.co/RichardErkhov/tokyotech-llm_-_Llama-3-Swallow-70B-v0.1-gguf/blob/main/Llama-3-Swallow-70B-v0.1.IQ4_XS.gguf) | IQ4_XS | 35.64GB |
26
+ | [Llama-3-Swallow-70B-v0.1.Q4_0.gguf](https://huggingface.co/RichardErkhov/tokyotech-llm_-_Llama-3-Swallow-70B-v0.1-gguf/blob/main/Llama-3-Swallow-70B-v0.1.Q4_0.gguf) | Q4_0 | 37.22GB |
27
+ | [Llama-3-Swallow-70B-v0.1.IQ4_NL.gguf](https://huggingface.co/RichardErkhov/tokyotech-llm_-_Llama-3-Swallow-70B-v0.1-gguf/tree/main/) | IQ4_NL | 37.58GB |
28
+ | [Llama-3-Swallow-70B-v0.1.Q4_K_S.gguf](https://huggingface.co/RichardErkhov/tokyotech-llm_-_Llama-3-Swallow-70B-v0.1-gguf/tree/main/) | Q4_K_S | 37.58GB |
29
+ | [Llama-3-Swallow-70B-v0.1.Q4_K.gguf](https://huggingface.co/RichardErkhov/tokyotech-llm_-_Llama-3-Swallow-70B-v0.1-gguf/tree/main/) | Q4_K | 39.6GB |
30
+ | [Llama-3-Swallow-70B-v0.1.Q4_K_M.gguf](https://huggingface.co/RichardErkhov/tokyotech-llm_-_Llama-3-Swallow-70B-v0.1-gguf/tree/main/) | Q4_K_M | 39.6GB |
31
+ | [Llama-3-Swallow-70B-v0.1.Q4_1.gguf](https://huggingface.co/RichardErkhov/tokyotech-llm_-_Llama-3-Swallow-70B-v0.1-gguf/tree/main/) | Q4_1 | 41.27GB |
32
+ | [Llama-3-Swallow-70B-v0.1.Q5_0.gguf](https://huggingface.co/RichardErkhov/tokyotech-llm_-_Llama-3-Swallow-70B-v0.1-gguf/tree/main/) | Q5_0 | 45.32GB |
33
+ | [Llama-3-Swallow-70B-v0.1.Q5_K_S.gguf](https://huggingface.co/RichardErkhov/tokyotech-llm_-_Llama-3-Swallow-70B-v0.1-gguf/tree/main/) | Q5_K_S | 45.32GB |
34
+ | [Llama-3-Swallow-70B-v0.1.Q5_K.gguf](https://huggingface.co/RichardErkhov/tokyotech-llm_-_Llama-3-Swallow-70B-v0.1-gguf/tree/main/) | Q5_K | 46.52GB |
35
+ | [Llama-3-Swallow-70B-v0.1.Q5_K_M.gguf](https://huggingface.co/RichardErkhov/tokyotech-llm_-_Llama-3-Swallow-70B-v0.1-gguf/tree/main/) | Q5_K_M | 46.52GB |
36
+ | [Llama-3-Swallow-70B-v0.1.Q5_1.gguf](https://huggingface.co/RichardErkhov/tokyotech-llm_-_Llama-3-Swallow-70B-v0.1-gguf/tree/main/) | Q5_1 | 49.36GB |
37
+ | [Llama-3-Swallow-70B-v0.1.Q6_K.gguf](https://huggingface.co/RichardErkhov/tokyotech-llm_-_Llama-3-Swallow-70B-v0.1-gguf/tree/main/) | Q6_K | 53.91GB |
38
+ | [Llama-3-Swallow-70B-v0.1.Q8_0.gguf](https://huggingface.co/RichardErkhov/tokyotech-llm_-_Llama-3-Swallow-70B-v0.1-gguf/tree/main/) | Q8_0 | 69.83GB |
39
+
40
+
41
+
42
+
43
+ Original model description:
44
+ ---
45
+ language:
46
+ - en
47
+ - ja
48
+ library_name: transformers
49
+ pipeline_tag: text-generation
50
+ license: llama3
51
+ model_type: llama
52
+ ---
53
+
54
+ # Llama3 Swallow
55
+
56
+ Our Swallow model has undergone continual pre-training from the [Llama 3 family](https://huggingface.co/collections/meta-llama/meta-llama-3-66214712577ca38149ebb2b6), primarily with the addition of Japanese language data. The Instruct versions use supervised fine-tuning (SFT) and Chat Vector. Links to other models can be found in the index.
57
+
58
+
59
+ # Model Release Updates
60
+
61
+ We are excited to share the release schedule for our latest models:
62
+ - **July 1, 2024**: Released the [Llama-3-Swallow-8B-v0.1](https://huggingface.co/tokyotech-llm/Llama-3-Swallow-8B-v0.1), [Llama-3-Swallow-8B-Instruct-v0.1](https://huggingface.co/tokyotech-llm/Llama-3-Swallow-8B-Instruct-v0.1), [Llama-3-Swallow-70B-v0.1](https://huggingface.co/tokyotech-llm/Llama-3-Swallow-70B-v0.1), and [Llama-3-Swallow-70B-Instruct-v0.1](https://huggingface.co/tokyotech-llm/Llama-3-Swallow-70B-Instruct-v0.1).
63
+
64
+ ## Swallow Model Index
65
+
66
+ |Model|Llama-3-Swallow|Llama3 Swallow Instruct|
67
+ |---|---|---|
68
+ |8B| [Link](https://huggingface.co/tokyotech-llm/Llama-3-Swallow-8B-v0.1) | [Link](https://huggingface.co/tokyotech-llm/Llama-3-Swallow-8B-Instruct-v0.1) |
69
+ |70B| [Link](https://huggingface.co/tokyotech-llm/Llama-3-Swallow-70B-v0.1) | [Link](https://huggingface.co/tokyotech-llm/Llama-3-Swallow-70B-Instruct-v0.1) |
70
+
71
+ ![logo](./logo.png)
72
+
73
+ This repository provides large language models developed by [Swallow-LLM](https://swallow-llm.github.io/).
74
+ Read our [blog post](https://zenn.dev/tokyotech_lm/articles/f65989d76baf2c).
75
+
76
+ ## Model Details
77
+
78
+ * **Model type**: Please refer to [Llama 3 MODEL_CARD](https://github.com/meta-llama/llama3/blob/main/MODEL_CARD.md) for details on the model architecture.
79
+ * **Language(s)**: Japanese English
80
+ * **Library**: [Megatron-LM](https://github.com/NVIDIA/Megatron-LM)
81
+ * **Tokenizer**: Please refer to [Llama 3 blog](https://ai.meta.com/blog/meta-llama-3/) for details on the tokenizer.
82
+ * **Contact**: swallow[at]nlp.c.titech.ac.jp
83
+
84
+ ## Model Performance
85
+
86
+ ### Japanese tasks
87
+
88
+ |Model|Size|JCom.|JEMHopQA|NIILC|JSQuAD|XL-Sum|MGSM|WMT20-en-ja|WMT20-ja-en|JMMLU|JHumanEval|Ja Avg|
89
+ |---|---|---|---|---|---|---|---|---|---|---|---|---|
90
+ | | |4-shot|4-shot|4-shot|4-shot|1-shot|4-shot|4-shot|4-shot|5-shot|0-shot| |
91
+ | | |EM acc|Char-F1|Char-F1|Char-F1|ROUGE-2|EM acc|BLEU|BLEU|EM acc|pass@1| |
92
+ |Llama-2-70b|70B|0.8651|0.5157|0.5464|0.9130|0.2372|0.3640|0.2657|0.2402|0.5496|0.2841|0.4781|
93
+ |Swallow-70b-hf|70B|0.9178|0.6178|**0.6910**|0.9208|0.2279|0.4720|0.3046|0.2301|0.5750|0.2262|0.5183|
94
+ |Qwen2-72B|72B|0.9607|0.6399|0.5617|**0.9261**|0.2362|**0.7560**|0.2747|0.2419|**0.7831**|**0.5567**|**0.5937**|
95
+ |Meta-Llama-3-70B|70B|0.9473|0.6042|0.5965|0.9207|0.2254|0.6720|0.2855|0.2526|0.6975|0.4799|0.5682|
96
+ |Llama-3-Swallow-70B-v0.1|70B|**0.9714**|**0.6695**|0.6881|0.9218|**0.2404**|0.7080|**0.3072**|**0.2548**|0.7049|0.4683|0.5934|
97
+
98
+ ### English tasks
99
+
100
+ |Model|Size|OpenBookQA|TriviaQA|HellaSWAG|SQuAD2.0|XWINO|MMLU|GSM8K|BBH|HumanEval|En Avg|
101
+ |---|---|---|---|---|---|---|---|---|---|---|---|
102
+ | | |4-shot|4-shot|4-shot|4-shot|4-shot|5-shot|4-shot|3-shot|0-shot| |
103
+ | | |Acc|EM acc|Acc|EM acc|Acc|Acc|EM acc|CoT EM Acc|pass@1| |
104
+ |Llama-2-70b|70B|0.4260|0.7988|0.6681|0.3379|**0.9256**|0.6876|0.5466|0.6643|0.3152|0.5967|
105
+ |Swallow-70b-hf|70B|0.4160|0.7610|0.6433|0.3345|0.9191|0.6571|0.5080|0.6537|0.2409|0.5704|
106
+ |Qwen2-72B|72B|0.4160|0.7890|0.6766|0.4052|0.9161|**0.8428**|**0.8908**|0.6388|**0.6049**|0.6867|
107
+ |Meta-Llama-3-70B|70B|**0.4360**|**0.8263**|**0.6909**|**0.4071**|0.9213|0.7870|0.8014|**0.8266**|0.5177|**0.6905**|
108
+ |Llama-3-Swallow-70B-v0.1|70B|0.4240|0.8231|0.6828|0.4059|0.9234|0.7745|0.8143|0.7352|0.4909|0.6749|
109
+
110
+ ## Evaluation Benchmarks
111
+
112
+ ### Japanese evaluation benchmarks
113
+
114
+ We used llm-jp-eval(v1.3.0), JP Language Model Evaluation Harness(commit #9b42d41) and Code Generation LM Evaluation Harness(commit #0261c52). The details are as follows:
115
+
116
+ - Multiple-choice question answering (JCommonsenseQA [Kurihara et al., 2022])
117
+ - Open-ended question answering (JEMHopQA [Ishii et al., 2024])
118
+ - Open-ended question answering (NIILC [関根, 2003])
119
+ - Machine reading comprehension (JSQuAD [Kurihara et al., 2022])
120
+ - Automatic summarization (XL-Sum [Hasan et al., 2021])
121
+ - Machine translation (WMT2020 ja-en [Barrault et al., 2020])
122
+ - Machine translation (WMT2020 en-ja [Barrault et al., 2020])
123
+ - Mathematical reasoning (MGSM [Shi et al., 2023])
124
+ - Academic exams (JMMLU [尹ら, 2024])
125
+ - Code generation (JHumanEval [佐藤ら, 2024])
126
+
127
+ ### English evaluation benchmarks
128
+
129
+ We used the Language Model Evaluation Harness(v.0.4.2) and Code Generation LM Evaluation Harness(commit #0261c52). The details are as follows:
130
+
131
+ - Multiple-choice question answering (OpenBookQA [Mihaylov et al., 2018])
132
+ - Open-ended question answering (TriviaQA [Joshi et al., 2017])
133
+ - Machine reading comprehension (SQuAD2 [Rajpurkar et al., 2018])
134
+ - Commonsense reasoning (XWINO [Tikhonov and Ryabinin, 2021])
135
+ - Natural language inference (HellaSwag [Zellers et al., 2019])
136
+ - Mathematical reasoning (GSM8K [Cobbe et al., 2021])
137
+ - Reasoning (BBH (BIG-Bench-Hard) [Suzgun et al., 2023])
138
+ - Academic exams (MMLU [Hendrycks et al., 2021])
139
+ - Code generation (HumanEval [Chen et al., 2021])
140
+
141
+ ## Training Datasets
142
+
143
+ ### Continual Pre-Training
144
+ The following datasets were used for continual pre-training.
145
+
146
+ - [Algebraic Stack](https://huggingface.co/datasets/EleutherAI/proof-pile-2)
147
+ - [Cosmopedia](https://huggingface.co/datasets/HuggingFaceTB/cosmopedia)
148
+ - [English Wikipedia](https://dumps.wikimedia.org/other/cirrussearch)
149
+ - [Japanese Wikipedia](https://dumps.wikimedia.org/other/cirrussearch)
150
+ - [Laboro ParaCorpus](https://github.com/laboroai/Laboro-ParaCorpus)
151
+ - [OpenWebMath](https://huggingface.co/datasets/EleutherAI/proof-pile-2)
152
+ - [RefinedWeb](https://huggingface.co/datasets/tiiuae/falcon-refinedweb)
153
+ - [Swallow Corpus](https://arxiv.org/abs/2404.17733)
154
+
155
+ ## Risks and Limitations
156
+
157
+ The models released here are still in the early stages of our research and development and have not been tuned to ensure outputs align with human intent and safety considerations.
158
+
159
+ ## Acknowledgements
160
+
161
+ We thank Meta Research for releasing Llama 3 under an open license for others to build on.
162
+
163
+ Our project is supported by the [Large Generative AI Development Support Program](https://abci.ai/en/link/lfm_support_program.html) of the National Institute of Advanced Industrial Science and Technology.
164
+
165
+ ## License
166
+
167
+ [META LLAMA 3 COMMUNITY LICENSE](https://llama.meta.com/llama3/license/)
168
+
169
+ ## Authors
170
+
171
+ Here are the team members:
172
+ - From [Tokyo Institute of Technology Okazaki Laboratory](https://www.nlp.c.titech.ac.jp/index.en.html), the following members:
173
+ - [Naoaki Okazaki](https://www.chokkan.org/index.ja.html)
174
+ - [Sakae Mizuki](https://s-mizuki-nlp.github.io/)
175
+ - [Youmi Ma](https://www.nlp.c.titech.ac.jp/member/youmi.en.html)
176
+ - [Koki Maeda](https://sites.google.com/view/silviase)
177
+ - [Kakeru Hattori](https://aya-se.vercel.app/)
178
+ - [Masanari Ohi](https://sites.google.com/view/masanariohi)
179
+ - [Taihei Shiotani](https://github.com/inatoihs)
180
+ - [Koshiro Saito](https://sites.google.com/view/koshiro-saito)
181
+ - From [Tokyo Institute of Technology YOKOTA Laboratory](https://www.rio.gsic.titech.ac.jp/en/index.html), the following members:
182
+ - [Rio Yokota](https://twitter.com/rioyokota)
183
+ - [Kazuki Fujii](https://twitter.com/okoge_kaz)
184
+ - [Taishi Nakamura](https://twitter.com/Setuna7777_2)
185
+ - [Takumi Okamoto](https://www.linkedin.com/in/takumi-okamoto)
186
+ - [Ishida Shigeki](https://www.wantedly.com/id/reborn27)
187
+ - From [Artificial Intelligence Research Center, AIST, Japan](https://www.airc.aist.go.jp/en/teams/), the following members:
188
+ - [Hiroya Takamura](https://sites.google.com/view/hjtakamura)
189
+
190
+ ## How to cite
191
+
192
+ If you find our work helpful, please feel free to cite us.
193
+
194
+ ```
195
+ @inproceedings{Fujii:COLM2024,
196
+ title={Continual Pre-Training for Cross-Lingual LLM Adaptation:
197
+ Enhancing Japanese Language Capabilities},
198
+ author={Kazuki Fujii and Taishi Nakamura and Mengsay Loem and Hiroki
199
+ Iida and Masanari Ohi and Kakeru Hattori and Hirai Shota and Sakae
200
+ Mizuki and Rio Yokota and Naoaki Okazaki},
201
+ booktitle="Proceedings of the First Conference on Language Modeling",
202
+ series={COLM},
203
+ pages="(to appear)",
204
+ year="2024",
205
+ month=oct,
206
+ address={University of Pennsylvania, USA},
207
+ }
208
+
209
+ @inproceedings{Okazaki:COLM2024,
210
+ title={Building a Large Japanese Web Corpus for Large Language Models},
211
+ author={Naoaki Okazaki and Kakeru Hattori and Hirai Shota and Hiroki
212
+ Iida and Masanari Ohi and Kazuki Fujii and Taishi Nakamura and Mengsay
213
+ Loem and Rio Yokota and Sakae Mizuki},
214
+ booktitle="Proceedings of the First Conference on Language Modeling",
215
+ series={COLM},
216
+ pages="(to appear)",
217
+ year="2024",
218
+ month=oct,
219
+ address={University of Pennsylvania, USA},
220
+ }
221
+ ```
222
+
223
+ ### Citations
224
+
225
+ ```tex
226
+ @article{llama3modelcard,
227
+ title={Llama 3 Model Card},
228
+ author={AI@Meta},
229
+ year={2024},
230
+ url = {https://github.com/meta-llama/llama3/blob/main/MODEL_CARD.md}
231
+ }
232
+ ```
233
+