polm commited on
Commit
4ae7fd9
1 Parent(s): 131acac

Add README etc.

Browse files
README.md CHANGED
@@ -1,3 +1,139 @@
1
  ---
2
- license: llama2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
+ language:
3
+ - ja
4
+ tags:
5
+ - japanese-stablelm
6
+ - causal-lm
7
+ pipeline_tag: text-generation
8
+ datasets:
9
+ - wikipedia
10
+ - mc4
11
+ - cc100
12
+ - oscar-corpus/OSCAR-2301
13
+ - oscar-corpus/OSCAR-2201
14
+ - cerebras/SlimPajama-627B
15
+ license:
16
+ - llama2
17
+ extra_gated_fields:
18
+ Name: text
19
+ Email: text
20
+ Country: text
21
+ Organization or Affiliation: text
22
+ I allow Stability AI to contact me about information related to its models and research: checkbox
23
  ---
24
+
25
+ # Japanese-StableLM-Base-Beta-70B
26
+
27
+ ![A cute robot wearing a kimono writes calligraphy with one single brush](./japanese-stablelm-robot.jpg)
28
+
29
+ > A cute robot wearing a kimono writes calligraphy with one single brush — [Stable Diffusion XL](https://clipdrop.co/stable-diffusion)
30
+
31
+ ## Model Description
32
+
33
+ `japanese-stablelm-base-beta-70b` is a 70B-parameter decoder-only language model based on [Llama-2-70b](https://huggingface.co/meta-llama/Llama-2-70b) that has been fine-tuned on a diverse collection of Japanese data, with the intent of maximizing downstream performance on Japanese language tasks.
34
+
35
+ For an instruction-following model, check [Japanese-StableLM-Instruct-Beta-70B](https://huggingface.co/stabilityai/japanese-stablelm-instruct-beta-70b). The base and instruct models are also available in smaller 7b sizes. For a model that has faster inference times, see [Japanese-StableLM-Base-JA_Vocab-Beta-7B](https://huggingface.co/stabilityai/japanese-stablelm-base-ja_vocab-beta-7b), or [the instruction-following version](https://huggingface.co/stabilityai/japanese-stablelm-instruct-ja_vocab-beta-7b).
36
+
37
+ ## Usage
38
+
39
+ First install additional dependencies in [requirements.txt](./requirements.txt):
40
+
41
+ ```sh
42
+ pip install -r requirements.txt
43
+ ```
44
+
45
+ Then start generating text with `japanese-stablelm-base-beta-70b` by using the following code snippet:
46
+
47
+ ```python
48
+ import torch
49
+ from transformers import AutoTokenizer, AutoModelForCausalLM
50
+
51
+ model_name = "stabilityai/japanese-stablelm-base-beta-70b"
52
+ tokenizer = AutoTokenizer.from_pretrained(model_name)
53
+
54
+ # The next line may need to be modified depending on the environment
55
+ model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype=torch.float16, low_cpu_mem_usage=True, device_map="auto")
56
+
57
+ prompt = """
58
+ AI で科学研究を加速するには、
59
+ """.strip()
60
+
61
+ input_ids = tokenizer.encode(
62
+ prompt,
63
+ add_special_tokens=False,
64
+ return_tensors="pt"
65
+ )
66
+
67
+ # this is for reproducibility.
68
+ # feel free to change to get different result
69
+ seed = 23
70
+ torch.manual_seed(seed)
71
+
72
+ tokens = model.generate(
73
+ input_ids.to(device=model.device),
74
+ max_new_tokens=128,
75
+ temperature=0.99,
76
+ top_p=0.95,
77
+ do_sample=True,
78
+ )
79
+
80
+ out = tokenizer.decode(tokens[0], skip_special_tokens=True)
81
+ print(out)
82
+ ```
83
+
84
+ We suggest playing with different generation config (`top_p`, `repetition_penalty` etc) to find the best setup for your tasks. For example, use higher temperature for roleplay task, lower temperature for reasoning.
85
+
86
+ ## Model Details
87
+
88
+ * **Model type**: `japanese-stablelm-base-beta-70b` model is an auto-regressive language model based on the Llama2 transformer architecture.
89
+ * **Language(s)**: Japanese
90
+ * **License**: [Llama2 Community License](https://ai.meta.com/llama/license/).
91
+ * **Contact**: For questions and comments about the model, please join [Stable Community Japan](https://discord.gg/StableJP). For future announcements / information about Stability AI models, research, and events, please follow https://twitter.com/StabilityAI_JP.
92
+
93
+ ## Training Dataset
94
+
95
+ Roughly 100B tokens from a mixture of the following corpora were used for continued pre-training.
96
+
97
+ - [Japanese/English Wikipedia](https://dumps.wikimedia.org/other/cirrussearch)
98
+ - [Japanese mc4](https://huggingface.co/datasets/mc4)
99
+ - [Japanese CC-100](http://data.statmt.org/cc-100/ja.txt.xz)
100
+ - [Japanese OSCAR](https://oscar-project.github.io/documentation/)
101
+ - [SlimPajama](https://huggingface.co/datasets/cerebras/SlimPajama-627B) (excluding the Books3 subset)
102
+
103
+ ## Use and Limitations
104
+
105
+ ### Intended Use
106
+
107
+ The model is intended to be used by all individuals as a foundation for application-specific fine-tuning without strict limitations on commercial use.
108
+
109
+ ### Limitations and bias
110
+
111
+ The pre-training dataset may have contained offensive or inappropriate content even after applying data cleansing filters which can be reflected in the model generated text. We recommend users exercise reasonable caution when using these models in production systems. Do not use the model for any applications that may cause harm or distress to individuals or groups.
112
+
113
+ ## Authors
114
+ This model was developed by the Research & Development team at Stability AI Japan, and the development was co-led by [Takuya Akiba](https://huggingface.co/iwiwi) and [Meng Lee](https://huggingface.co/leemeng). The members of the team are as follows:
115
+
116
+ - [Meng Lee](https://huggingface.co/leemeng)
117
+ - [Fujiki Nakamura](https://huggingface.co/fujiki)
118
+ - [Makoto Shing](https://huggingface.co/mkshing)
119
+ - [Paul McCann](https://huggingface.co/polm-stability)
120
+ - [Takuya Akiba](https://huggingface.co/iwiwi)
121
+ - [Naoki Orii](https://huggingface.co/mrorii)
122
+
123
+ ## Acknowledgements
124
+
125
+ We thank Meta Research for releasing Llama 2 under an open license for others to build on.
126
+
127
+ We are grateful for the contributions of the EleutherAI Polyglot-JA team in helping us to collect a large amount of pre-training data in Japanese. Polyglot-JA members includes Hyunwoong Ko (Project Lead), Fujiki Nakamura (originally started this project when he commited to the Polyglot team), Yunho Mo, Minji Jung, KeunSeok Im, and Su-Kyeong Jang.
128
+
129
+ We are also appreciative of [AI Novelist/Sta (Bit192, Inc.)](https://ai-novel.com/index.php) and the numerous contributors from [Stable Community Japan](https://discord.gg/VPrcE475HB) for assisting us in gathering a large amount of high-quality Japanese textual data for model training.
130
+
131
+ ## How to cite
132
+ ```
133
+ @misc{JapaneseStableLMBaseBeta70B,
134
+ url={[https://huggingface.co/stabilityai/japanese-stablelm-base-beta-70b](https://huggingface.co/stabilityai/japanese-stablelm-base-beta-70b)},
135
+ title={Japanese StableLM Base Beta 70B},
136
+ author={Lee, Meng and Nakamura, Fujiki and Shing, Makoto and McCann, Paul and Akiba, Takuya and Orii, Naoki}
137
+ }
138
+ ```
139
+
config.json ADDED
@@ -0,0 +1,25 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "LlamaForCausalLM"
4
+ ],
5
+ "bos_token_id": 1,
6
+ "eos_token_id": 2,
7
+ "hidden_act": "silu",
8
+ "hidden_size": 8192,
9
+ "initializer_range": 0.02,
10
+ "intermediate_size": 28672,
11
+ "max_position_embeddings": 4096,
12
+ "model_type": "llama",
13
+ "num_attention_heads": 64,
14
+ "num_hidden_layers": 80,
15
+ "num_key_value_heads": 8,
16
+ "pad_token_id": 0,
17
+ "pretraining_tp": 1,
18
+ "rms_norm_eps": 1e-05,
19
+ "rope_scaling": null,
20
+ "tie_word_embeddings": false,
21
+ "torch_dtype": "float16",
22
+ "transformers_version": "4.31.0",
23
+ "use_cache": true,
24
+ "vocab_size": 32000
25
+ }
generation_config.json ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "bos_token_id": 1,
4
+ "eos_token_id": 2,
5
+ "pad_token_id": 0,
6
+ "transformers_version": "4.31.0"
7
+ }
japanese-stablelm-robot.jpg ADDED
requirements.txt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ sentencepiece
2
+ protobuf
3
+ accelerate