TTP01 commited on
Commit
6b4e4a8
·
verified ·
1 Parent(s): 01f0c51

Upload README.md

Browse files
Files changed (1) hide show
  1. README.md +27 -8
README.md CHANGED
@@ -70,8 +70,8 @@ configs:
70
 
71
  | 🌐 Ngôn Ngữ / Language | 📌 Chuyển Hướng / Quick Navigation |
72
  | :--- | :--- |
73
- | **🇻🇳 Bản Tiếng Việt Đầy Đủ** | 👉 [**Xem toàn bộ Báo cáo Kỹ thuật Tiếng Việt**](#vietnamese) |
74
- | **🇬🇧 Full English Version** | 👉 [**Jump to the complete English Technical Dataset Card**](#english) |
75
 
76
  </div>
77
 
@@ -420,7 +420,25 @@ trainer.train()
420
 
421
  ---
422
 
423
- ## 8. Ủng Hộ Dự Án & Liên Hệ (Support & Contact)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
424
 
425
  Nếu bạn thấy dự án **Vietverse-SFT** hữu ích và muốn tiếp thêm nguồn lực (chi phí API & Compute) để mở rộng các tập dữ liệu tiếng Việt quy mô lớn hơn trong tương lai, mọi sự ủng hộ đều rất quý báu:
426
 
@@ -574,6 +592,7 @@ dataset = load_dataset("TTP01/Vietverse-SFT-Preview", split="train")
574
  print(f"✅ Loaded {len(dataset)} samples")
575
  print("User:", dataset[0]["messages"][0]["content"])
576
  print("Assistant:", dataset[0]["messages"][1]["content"])
 
577
  ```
578
 
579
  ---
@@ -606,11 +625,11 @@ model = FastLanguageModel.get_peft_model(
606
  # 3. Format Chat Template
607
  dataset = load_dataset("TTP01/Vietverse-SFT-Preview", split="train")
608
 
609
- def format_prompts(batch):
610
  texts = [tokenizer.apply_chat_template(msgs, tokenize=False, add_generation_prompt=False) for msgs in batch["messages"]]
611
  return {"text": texts}
612
 
613
- dataset = dataset.map(format_prompts, batched=True)
614
 
615
  # 4. Train Model
616
  trainer = SFTTrainer(
@@ -766,10 +785,10 @@ trainer.train()
766
 
767
  ---
768
 
769
- ## 7. Giới Hạn & Đạo Đức Dữ Liệu (Limitations & Ethics)
770
 
771
- - **Dữ liệu tổng hợp (Synthetic Data Boundary)**: đã qua 3 tầng lọc và kiểm chứng G-Eval, dữ liệu tổng hợp từ LLM vẫn thể tồn tại sai sót nhỏ ở các mốc thời gian hoặc số hiệu văn bản pháp lý rất chi tiết. Chúng tôi khuyến nghị người dùng đối chiếu văn bản gốc khi áp dụng cho các quyết định pháp lý và y khoa quan trọng.
772
- - **Tính an toàn**: Tập dữ liệu đã được lọc sạch các nội dung độc hại, vi phạm pháp luật đảm bảo tính bảo mật cá nhân (No PII).
773
 
774
  ---
775
 
 
70
 
71
  | 🌐 Ngôn Ngữ / Language | 📌 Chuyển Hướng / Quick Navigation |
72
  | :--- | :--- |
73
+ | **🇻🇳 Bản Tiếng Việt Đầy Đủ (9 Mục)** | 👉 [**Xem toàn bộ Báo cáo Kỹ thuật Tiếng Việt**](#vietnamese) |
74
+ | **🇬🇧 Full English Version (9 Sections)** | 👉 [**Jump to the complete English Technical Dataset Card**](#english) |
75
 
76
  </div>
77
 
 
420
 
421
  ---
422
 
423
+ ## 8. Giấy Phép & Trích Dẫn (License & Citation)
424
+
425
+ - **Giấy phép**: Phát hành theo giấy phép mở **Apache License 2.0**.
426
+ - **Trích dẫn BibTeX**:
427
+
428
+ ```bibtex
429
+ @dataset{vietverse_sft_preview_2026,
430
+ author = {TTP01 and Vietverse Contributors},
431
+ title = {Vietverse-SFT: A High-Density, Native Vietnamese Foundation Dataset for Large Language Model Instruction Tuning},
432
+ year = {2026},
433
+ publisher = {Hugging Face},
434
+ howpublished = {\url{https://huggingface.co/datasets/TTP01/Vietverse-SFT-Preview}},
435
+ version = {1.0-preview}
436
+ }
437
+ ```
438
+
439
+ ---
440
+
441
+ ## 9. Ủng Hộ Dự Án & Liên Hệ (Support & Contact)
442
 
443
  Nếu bạn thấy dự án **Vietverse-SFT** hữu ích và muốn tiếp thêm nguồn lực (chi phí API & Compute) để mở rộng các tập dữ liệu tiếng Việt quy mô lớn hơn trong tương lai, mọi sự ủng hộ đều rất quý báu:
444
 
 
592
  print(f"✅ Loaded {len(dataset)} samples")
593
  print("User:", dataset[0]["messages"][0]["content"])
594
  print("Assistant:", dataset[0]["messages"][1]["content"])
595
+ print("Quality Score:", dataset[0]["quality_score"])
596
  ```
597
 
598
  ---
 
625
  # 3. Format Chat Template
626
  dataset = load_dataset("TTP01/Vietverse-SFT-Preview", split="train")
627
 
628
+ def apply_template(batch):
629
  texts = [tokenizer.apply_chat_template(msgs, tokenize=False, add_generation_prompt=False) for msgs in batch["messages"]]
630
  return {"text": texts}
631
 
632
+ dataset = dataset.map(apply_template, batched=True)
633
 
634
  # 4. Train Model
635
  trainer = SFTTrainer(
 
785
 
786
  ---
787
 
788
+ ## 7. Limitations & Ethical Considerations
789
 
790
+ - **Synthetic Data Boundary**: While audited through 3-stage multi-agent verification, synthetic generations may occasionally contain minor factual inaccuracies in niche legal citations or dates. Users should cross-reference original primary sources for critical legal or medical decisions.
791
+ - **Safety & PII**: The corpus has been strictly filtered and decontaminated against personally identifiable information (PII), toxic expressions, and sensitive private data.
792
 
793
  ---
794