metadata

language:
  - uz
license: apache-2.0
size_categories:
  - 100K<n<1M
task_categories:
  - question-answering
  - translation
pretty_name: Open Assistant Conversations Dataset Release 2 in Uzbek
dataset_info:
  features:
    - name: message_id
      dtype: string
    - name: parent_id
      dtype: string
    - name: user_id
      dtype: string
    - name: created_date
      dtype: string
    - name: text
      dtype: string
    - name: role
      dtype: string
    - name: lang
      dtype: string
    - name: review_count
      dtype: int64
    - name: review_result
      dtype: bool
    - name: deleted
      dtype: bool
    - name: rank
      dtype: float64
    - name: synthetic
      dtype: bool
    - name: model_name
      dtype: 'null'
    - name: detoxify
      struct:
        - name: identity_attack
          dtype: float64
        - name: insult
          dtype: float64
        - name: obscene
          dtype: float64
        - name: severe_toxicity
          dtype: float64
        - name: sexual_explicit
          dtype: float64
        - name: threat
          dtype: float64
        - name: toxicity
          dtype: float64
    - name: message_tree_id
      dtype: string
    - name: tree_state
      dtype: string
    - name: emojis
      struct:
        - name: count
          sequence: int64
        - name: name
          sequence: string
    - name: labels
      struct:
        - name: count
          sequence: int64
        - name: name
          sequence: string
        - name: value
          sequence: float64
  splits:
    - name: validation
      num_bytes: 5261768
      num_examples: 5123
    - name: train
      num_bytes: 128642107
      num_examples: 125181
  download_size: 43601021
  dataset_size: 133903875
configs:
  - config_name: default
    data_files:
      - split: validation
        path: data/validation-*
      - split: train
        path: data/train-*
tags:
  - human-feedback

Open Assistant Conversations Dataset Release 2 (OASST2) in Uzbek language

This dataset is an Uzbek translated version of OASST2 dataset.

Llama3 chat template + thread formatted dataset based on this translation is also available for model fine-tuning here.

The Uzbek translation was completed in 45 hours using a single T4 GPU and nllb-200-3.3B model.

Based on nllb metrics, you might want to only filter out records that were not originally in English or Russian since English-Uzbek and Russian-Uzbek have acceptable metrics and translation quality is noticeable better for those pairs based on my short reviews.

I am sharing the entire Uzbek translated dataset for future research.

The following repo and command was used to do the Uzbek translation.

Repo: https://github.com/UnderstandLingBV/LLaMa2lang

Command used:

!python3 translate.py nllb --model_size 3.3B uzn_Latn output_uzbek --quant8 --base_dataset OpenAssistant/oasst2 --max_length 512 --checkpoint_n 400 --batch_size 40

I will fine-tune LLAMA3 8B Uzbek chat model and release in HF soon.