--- language: ja tags: - t5 - text2text-generation - pilota license: apache-2.0 --- # Pilota model for dialogs A model for [Pilota](https://github.com/megagonlabs/pilota) trained with [Accommodation Search Dialog Corpus](https://github.com/megagonlabs/asdc) and other additional examples - ``scud`` - Fine tuned model of [t5-base-japanese-web (with Byte-fallback, 8K)](https://huggingface.co/megagonlabs/t5-base-japanese-web-8k) - The original model is distributed in [the Apache License 2.0](https://www.apache.org/licenses/LICENSE-2.0) - ``scorer`` - Fine tuned model of [LINE DistilBERT Japanese](https://huggingface.co/line-corporation/line-distilbert-base-japanese) - The original model is distributed in [the Apache License 2.0](https://www.apache.org/licenses/LICENSE-2.0) ## Usage 1. Install [Pilota](https://github.com/megagonlabs/pilota) 2. Prepare inputs - Command ```bash echo -e 'ご要望をお知らせください\tはい。部屋から富士山が見えて、夜景を見ながら食事のできるホテルがいいな。\nこんにちは\tこんにちは' | python -m pilota.convert.plain2request | tee input.jsonl ``` - Output ```jsonl {"context": [{"name": "agent", "text": "ご要望をお知らせください"}], "utterance": "はい。部屋から富士山が見えて、夜景を見ながら食事のできるホテルがいいな。", "sentences": null, "meta": {}} {"context": [{"name": "agent", "text": "こんにちは"}], "utterance": "こんにちは", "sentences": null, "meta": {}} ``` 3. Feed it to Pilota - Command ```console pilota -m megagonlabs/pilota_dialog --batch_size 1 --outlen 60 --nbest 1 --beam 5 < input.jsonl ``` - Output ```jsonl [{"scuds_nbest": [[]], "original_ranks": [0], "scores": [0.9911208689212798], "scores_detail": [{"OK": 0.9704028964042664, "incorrect_none": 0.04205145686864853, "lack": 0.0007874675211496651, "limited": 0.0003119863977190107, "non_fluent": 0.0002362923405598849, "untruth": 0.0013080810895189643}], "sentence": "はい。"}, {"scuds_nbest": [["部屋から富士山が見えるホテルが良い。", "夜景を見ながら食事のできるホテルが良い。"]], "original_ranks": [0], "scores": [0.9952289938926696], "scores_detail": [{"OK": 0.9840966463088989, "incorrect_none": 0.010280555114150047, "lack": 0.0032871251460164785, "limited": 0.00041511686868034303, "non_fluent": 0.0002954243100248277, "untruth": 0.003289491171017289}], "sentence": "部屋から富士山が見えて、夜景を見ながら食事のできるホテルがいいな。"}] [{"scuds_nbest": [[]], "original_ranks": [0], "scores": [0.9831213414669036], "scores_detail": [{"OK": 0.9704028964042664, "incorrect_none": 0.04205145686864853, "lack": 0.0007874675211496651, "limited": 0.0003119863977190107, "non_fluent": 0.0002362923405598849, "untruth": 0.0013080810895189643}], "sentence": "こんにちは"}] ``` ## License Apache License 2.0