Dataset Viewer
The dataset could not be loaded because the splits use different data file formats, which is not supported. Read more about the splits configuration. Click for more details.
Couldn't infer the same data file format for all splits. Got {NamedSplit('train'): (None, {}), NamedSplit('test'): ('json', {})}
Error code:   FileFormatMismatchBetweenSplitsError

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

PK-Genesis v1.1 Improvement Dataset

This dataset is separated into its own repository for iterative tuning.

Goal

Improve behavior on noisy/typo symptom inputs and prevent identity fallback on non-identity medical prompts.

File

  • pk_genesis_v1_1_improvement_dataset.jsonl

Format

{"prompt":"...","response":"...","category":"..."}

Key categories

  • typo-robustness
  • clarify-medical-typo
  • medical-nonidentity-control
  • identity-company-control
  • anti-identity-fallback

v1.2 healthcare performance set

  • File: pk_genesis_v1_2_healthcare_performance_dataset.jsonl
  • Rows: 940
  • Focus: typo robustness, crisis response quality, step-by-step intake, emergency triage consistency

Evaluation artifact

  • File: pkgenesis_health_eval_report.json
  • Key findings before v1.2 patch: identity leak on typo headance; mental-health crisis response needed richer safety guidance.

v1.3 typo + empathy hardening set

  • File: pk_genesis_v1_3_typo_empathy_dataset.jsonl
  • Rows: 820
  • Focus: misspelling robustness, empathetic bedside tone, anti-identity fallback on noisy symptom text (especially headance).

v1.4 Khmer healthcare set

  • File: pk_genesis_v1_4_khmer_healthcare_dataset.jsonl
  • Rows: 940
  • Focus: natural Khmer medical responses, emergency triage in Khmer, typo/noisy Khmer symptom handling, step-by-step intake in Khmer, and identity/company behavior only when asked.

v1.5 Khmer advanced healthcare set

  • File: pk_genesis_v1_5_khmer_advanced_dataset.jsonl
  • Rows: 1000
  • Focus: natural Khmer fluency (formal + colloquial), special populations (pediatric/pregnancy/elderly), medication safety phrasing, richer typo handling, and triage clarity.

Khmer evaluation artifact

  • File: pkgenesis_khmer_eval_report_v14.json
  • v1.4 eval summary: no identity leaks in tested Khmer set; emergency and multi-turn checks passed.

Khmer evaluation (v1.4 adapter)

  • File: pkgenesis_khmer_eval_report_v14.json
  • Result: no identity leaks, no emergency triage failures, no multi-turn failures in Khmer test set.

v1.5 Khmer conversation expansion

  • File: pk_genesis_v1_5_khmer_conversation_dataset.jsonl
  • Rows: 800
  • Focus: more natural Khmer conversational follow-up, empathetic tone, and triage clarity.

v1.6 consolidated all-dataset set

  • File: pk_genesis_v1_6_all_in_dataset.jsonl
  • Rows: 5280
  • Purpose: one combined dataset containing all prior PK-Genesis datasets for full consolidation finetune.
Downloads last month
6