StarsMakeGalaxy/prometheus-bfsi-tier1-triage
Pioneer BFSI / Fintech Tier-1 Autonomous Triage Dataset This dataset contains high-grade multi-turn conversational traces (ChatML format) curated according to the Pioneer paper data curation methodology for training and evaluating specialized 8B Small Language Models (SLMs) in the Banking, Financial Services, and Insurance (BFSI) vertical. Dataset Structure Train Split (train): 350 traces 75% Gold Standard Tasks: Standard operational workflows across 30… See the full description on the dataset page: https://huggingface.co/datasets/StarsMakeGalaxy/prometheus-bfsi-tier1-triage.
Pioneer BFSI / Fintech Tier-1 Autonomous Triage Dataset
This dataset contains high-grade multi-turn conversational traces (ChatML format) curated according to the Pioneer paper data curation methodology for training and evaluating specialized 8B Small Language Models (SLMs) in the Banking, Financial Services, and Insurance (BFSI) vertical.
Dataset Structure
- Train Split (`train`): 350 traces
- 75% Gold Standard Tasks: Standard operational workflows across 30 regulatory and banking intents with 4-step Chain-of-Thought (
<Thinking>) reasoning. - 25% Hard-Negative Disambiguation Pairs: Contrastive edge-case pairs resolving semantic ambiguities (e.g.
report_fraud_cardvscard_stolen_lost,stop_recurring_paymentvschargeback_dispute,transfer_stuck_pendingvsunauthorized_transfer). - Test Split (`test`): 100 frozen benchmark traces with 0.00% prompt contamination.
4-Step Chain-of-Thought <Thinking> Supervision
Every assistant turn features explicit multi-step reasoning:
- Input & Urgency Assessment (P0 Critical to P3 Routine)
- Entity & Parameter Extraction (amounts, transaction IDs, reference numbers)
- Policy & Regulatory Disambiguation (Regulation E, ERISA 402(c), SWIFT, KYC/AML)
- Operational Decision (Standardized Intent + Structured JSON Tool Payload + Customer Communication)
Dual-Output Schema
<Thinking>
1. Input & Urgency: ...
2. Entity & Parameter Extraction: ...
3. Policy & Regulatory Disambiguation: ...
4. Operational Decision: ...
</Thinking>
Intent: <intent_name>
Action: {"tool": "<tool_name>", "params": {...}}
Response: <customer_message>Intended Use with Unsloth
Format is directly compatible with Unsloth multi-turn conversational SFT:
from unsloth import FastLanguageModel
from datasets import load_dataset
dataset = load_dataset("StarsMakeGalaxy/prometheus-bfsi-tier1-triage")
# Apply chat template: get_chat_template(tokenizer, chat_template="chatml")