Align README format with core 4 model cards

README.md

@@ -6,26 +6,30 @@ library_name: transformers
 tags:
 - qwen3
 - english
-- data-science
 - data-mining
-- insight
+- data-science
+- instruction-tuning
 - sft
+- insight
 datasets:
 - zero9tech/data-scientist-insight-dialog-en-16.5k
 ---
 
-# QWEN3-4B-Data-Science-V2 - Insight Navigator
+# QWEN3-4B-Data-Science-V2
+
+This model is tuned for decision-oriented data mining and applied data science assistance.
 
 ## Training Setup
-Domain SFT on zero9tech/data-scientist-insight-dialog-en-16.5k.
+1. Domain SFT: zero9tech/data-scientist-insight-dialog-en-16.5k.
 
-## Dataset Snapshot
+## Dataset Test Highlights (zero9tech/data-scientist-insight-dialog-en-16.5k)
 - Total records: 16,463
-- Split: train 14,021 ; validation 801 ; test 1,641
-- Hard gate: PASS
-- Soft gate: PASS
-- Self-BLEU (final): 0.5574
-- Distinct-2 (final): 0.0768
+- Split: train: 14,021 · validation: 801 · test: 1,641
+- assistant_first_unique_ratio: 0.8408
+- assistant_final_unique_ratio: 1.0000
+
+## Usage Note
+Model behavior is optimized for decision-focused responses (method choice, alternatives, risk signals, validation planning).
 
 ## Copyright
 Copyright (c) Zero9 Tech