✕
⚡ NicTec JVPro — Login
Username
Password
Email
(for verification)
LOGIN
REGISTER
⬡ DROP FILES HERE
NicTec JVPro
v5.2 — Dataset Forge
Schema:
instruction, input, output
UI mode
…
EN
LOGOUT
📂 Select
🔍 Analyze
✅ Fix & Save
🔗 Merge
📊 Report
📄 Log
🕐 History
🗑 Reset
0
Files
0
Objects
0
Valid
0
Keys+
0
Types~
0
Short
0
Extra
0
Dupes
0
Removed
0
Plan C
📁 Files
⬡
Drop JSON files here
or click to select
Supported: .json, .jsonl
⚙ Schema
instruction input output
Alpaca
ShareGPT
Llama-3
✔ Apply Schema
🔒
Run
Analyze
to unlock
🧹 Strip Patterns
[?]=digits [word]=word [*]=any
Active patterns:
+ Add
✕ Del
🗺 Key Mapping
Source key → Target
→
+
✕
desc → instruction
input_text → input
result → output
— %
GOLD
Ideal for SFT
Confidence Score
◆ Gold
0
◇ Silver
0
◈ Bronze
0
✕ Trash
0
Gold
0–8 pts Ideal SFT
Silver
9–25 pts Pre-train
Bronze
26–60 pts Review
Trash
>60 pts Salvaged
Pipeline v5.2
KeyMap → Schema → Polish
📋 QUEUE
📊 STATS
📁 Queue
(0)
Show:
All
Done
Errors
No files
LOG:
Select a file from queue
FILE SUMMARY
Total objects
—
Valid
—
Dupes
—
Removed
—
Grade
—
Schema
—
🔒
Run Analyze
Patterns found
↻
—
+
✕
🔍
Analysis Result
Objects
—
Repair stages
—
Will remove
—
Dupes
—
Success rate
—
Exp. grade
—
Output lengths
👁 Preview changes
✅
Fix Result
Objects in
—
Objects out
—
Final grade
—
Saved
—
READY TO DOWNLOAD
⬇ СКАЧАТЬ РЕЗУЛЬТАТ
⚙ Processing
✨ Clean schema
Remove extra keys
🔁 Dedup (MD5)
Exact by instruction
≈ Fuzzy dedup
Semantic duplicates
🗺 Key Mapping
Auto + user-defined
🛡 Win Path Fix
C:\path\ patterns only
🧼 Deep Clean
mojibake · homoglyphs · entities · invisible chars
🕵 PII Masking
emails/phones/IPs → [EMAIL] [PHONE] [IP]
📦 Output
= input
JSONL
JSON
CSV
TSV
📏 Min output
chars
Порог схожести
0.75
🧠 LLM Mode
Enable LLM Polish
• Quotes [cite] / footnotes [42]
• Fences ``` (inline `code` safe ✓)
• Whitespace artifacts
Preset:
Alpaca Clean
ShareGPT Strict
Minimal (citations only)
Aggressive (all MD)
💾 Output
🔗 Merge files
→ merged_dataset.json
Merge & Export
merged_dataset_gold.json
⬡
Export
📋 Current Job
Grade:
—
Objects in:
0
Objects out:
0
Removed:
0
Schema:
—
Mappings:
0
Patterns:
0
🏅 Grade Health
◆
Gold
0
Ideal SFT
0–8 pts
◇
Silver
0
Pre-train
9–25 pts
◈
Bronze
0
Review
26–60 pts
✕
Trash
0
Salvaged
>60 pts
Ready
NicTec JVPro v5