Text Generation
Transformers
Safetensors
English
Japanese
llama
conversational
text-generation-inference
s-mizuki-nlp commited on
Commit
fdd4282
·
verified ·
1 Parent(s): 60ffc3e

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +9 -7
README.md CHANGED
@@ -184,17 +184,19 @@ print(output[0].outputs[0].text)
184
  The following instruction datasets were used for the instruction tuning.
185
  - Japanese
186
  - `lmsys-chat-1m-synth-ja-wo-pii-and-template-instructions`
187
- - Single-turn Japanese synthetic instruction dataset derived from [lmsys-chat-1m](https://huggingface.co/datasets/lmsys/lmsys-chat-1m) dataset [\[Zhang+, ICLR24\]](https://openreview.net/forum?id=BOfDKxfwt0)). The user instructions were translated into Japanese via machine translation, and the assistant responses were generated using the Llama 3.1 405B Instruct model. Rejection sampling was applied, with Llama 3.1 70B Instruct serving as the judge.
188
- - As implied by the dataset name, conversations that contain personally identifiable information (PII) or template-based user instructions have been removed.
189
- - filtered magpie-ultra-ja
190
- - Subset of the [magpie-ultra](https://huggingface.co/datasets/argilla/magpie-ultra-v0.1) dataset, containing samples rated as 'average,' 'good,' or 'excellent.'.
191
- - gemma-magpie
 
192
  - Generated using prompts for specific category words.
193
  - English
194
  - `lmsys-chat-1m-synth-en-wo-pii-and-template-instructions`
195
  - Similar to the `lmsys-chat-1m-synth-ja-wo-pii-and-template-instructions`, but this version uses the original English user instructions. The assistant responses were generated in English as well. Rejection sampling was not applied in this version.
196
- - filtered-magpie-ultra-en
197
- - Subset of the [magpie-ultra](https://huggingface.co/datasets/argilla/magpie-ultra-v0.1) dataset, containing samples rated as 'average,' 'good,' or 'excellent.'.
 
198
 
199
  ## Risks and Limitations
200
 
 
184
  The following instruction datasets were used for the instruction tuning.
185
  - Japanese
186
  - `lmsys-chat-1m-synth-ja-wo-pii-and-template-instructions`
187
+ - Single-turn Japanese synthetic instruction dataset derived from [lmsys-chat-1m](https://huggingface.co/datasets/lmsys/lmsys-chat-1m) dataset [\[Zhang+, ICLR24\]](https://openreview.net/forum?id=BOfDKxfwt0)).
188
+ The first-turn user instructions were translated into Japanese via DeepL machine translation, and the assistant responses were generated using the Llama 3.1 405B Instruct model. Rejection sampling (n=6) was applied, with Llama 3.1 70B Instruct serving as a judge.
189
+ - As implied by the dataset name, conversations that contain personally identifiable information (PII) or template-based user instructions have been removed. Duplicate instuctions have also been removed.
190
+ - `filtered-magpie-ultra-ja`
191
+ - A Japanese variant of the `filtered-magpie-ultra-en` dataset, machine-translated into Japanese using the Gemma 2 27B IT.
192
+ - `gemma-magpie`
193
  - Generated using prompts for specific category words.
194
  - English
195
  - `lmsys-chat-1m-synth-en-wo-pii-and-template-instructions`
196
  - Similar to the `lmsys-chat-1m-synth-ja-wo-pii-and-template-instructions`, but this version uses the original English user instructions. The assistant responses were generated in English as well. Rejection sampling was not applied in this version.
197
+ - `filtered-magpie-ultra-en`
198
+ - A subset of the [magpie-ultra](https://huggingface.co/datasets/argilla/magpie-ultra-v0.1) dataset, developed following the MAGPIE recipe [\[Xu+, arXiv24\]](https://arxiv.org/abs/2406.08464) using Llama 3.1 405B Instruct. This subset includes only samples rated as 'average,' 'good,' or 'excellent.'
199
+
200
 
201
  ## Risks and Limitations
202