Model Card for Model lianghsun/Llama-3.2-Taiwan-Legal-3B-Instruct
ๅบๆผ lianghsun/Llama-3.2-Taiwan-3B ๆจกๅ๏ผ้้ไธญ่ฏๆฐๅๅฐ็ฃๆณๅพๆขๆๅๅคๆฑบๆธ็ญ็ธ้่ณๆ้้ฒ่กๅพฎ่ชฟใ
Model Update History
Update Date | Model Version | Key Changes |
---|---|---|
2024-11-13 | v2024.11.13 | Fine-tuned version v2024.11.13 of lianghsun/Llama-3.2-Taiwan-3B. This instruction version initiates experimental integration of non-Chinese instructions to enhance the modelโs robustness and mitigate risks of overfitting. |
2024-11-06 | v2024.11.6 | Starting with this release, fine-tuning is based on the foundation model lianghsun/Llama-3.2-Taiwan-3B v2024.10.27 , and versioning has been updated to use the YYYY-mm-dd format. |
2024-10-17 | v1.1.0 (v2024.10.17) | (Model collapsed ๐ฅ) Experimental fine-tuning on v1.0.0 with added legal code data from the Republic of China (Taiwan) |
2024-10-10 | v1.0.0 (v2024.10.10) | Full model training completed, but missing legal code data for the Republic of China (Taiwan) |
2024-09-27 | v0.1.0 (v2024.09.27) | Model v0.1.0 released, but training was interrupted after 3 epochs due to lack of compute resources |
Model Details
Model Description
ๅบๆผ lianghsun/Llama-3.2-Taiwan-3B ๆจกๅ๏ผๆญคๅพฎ่ชฟ้็จไฝฟ็จไบไพ่ชไธญ่ฏๆฐๅๅฐ็ฃ็ๆณๅพๆขๆ่็ธ้ๅคๆฑบๆธ่ณๆ้๏ผไปฅๆๅๆจกๅๅจๆณๅพ้ ๅ็ๅฐๆฅญ็ฅ่ญ่ๆ็จ่ฝๅใ้ไบ่ณๆ้ๆถต่ไบๆณๅพๆขๆ็็ตๆงใๅคๆฑบๆธ็ๆ ผๅผ๏ผๆณๅบญไธๅธธ่ฆ็ๆณๅพ่ช่จ่่ก่ช๏ผไธฆๅ ๅซไบ้จๅๆณๅพ่ณๆ็งๅญธไปปๅ็ๆ็จ๏ผไฝฟๆจกๅ่ฝๅค ๆดๆบ็ขบๅฐ็่งฃๅ่็่ๅฐ็ฃๆณๅพ้ซ็ณป็ธ้็ๅ้กใ็ถ้้ไบๅพฎ่ชฟ๏ผๆจกๅๅฐ่ฝๅค ๆดๅฅฝๅฐ็บๆณๅพๅฐๆฅญไบบๅฃซๆไพๅนซๅฉ๏ผไธฆๅจๅฐ็ฃๆณๅถๆกๆถๅ งๆไพๆด็ฒพๆบ็ๅๆ่ๅปบ่ญฐใ
- Developed by: Huang Liang Hsun
- Model type: LlamaForCausalLM
- Language(s) (NLP): ไธป่ฆ่็็น้ซไธญๆ๏ผzh-tw๏ผ๏ผ้ๅฐไธญ่ฏๆฐๅๅฐ็ฃ็ๆณๅพ็จ่ช่ๅคๆฑบๆธ้ฒ่กๅพฎ่ชฟใ
- License: llama3.2
- Finetuned from model: lianghsun/Llama-3.2-Taiwan-3B
Model Sources
- Repository: lianghsun/Llama-3.2-Taiwan-Legal-3B-Instruct
- Demo: (WIP)
Uses
Direct Use
ๆญคๆจกๅๅฏไปฅ็ดๆฅ็จๆผ็่งฃๅ็ๆ็น้ซไธญๆๆณๅพๆๆฌ๏ผ้ฉๅ้่ฆ่็ๅฐ็ฃๆณๅพ็ธ้ๅ้ก็ๆ็จๅ ดๆฏใๆจกๅ้ ่จญ็ๆไปคๅๅๆ่ฝๅค ๆๆๆไพๆณๅพ่ณ่จใ้ๆธ ๆณๅพๆขๆใไธฆ็ๆ็ฌฆๅๆณๅพๅฐๆฅญ็ๅๆใๅ ถ็ดๆฅไฝฟ็จ็ฏๅๅ ๆฌไฝไธ้ๆผๆณๅพ่ณ่จๆฅ่ฉขใๆณๅพๆๆฌๆ่ฆใๅๅบๆฌ็ๆณๆขๅฐ่ฉฑใ
Downstream Use
็ถ้ๅพฎ่ชฟๅพ๏ผ่ฉฒๆจกๅๅฏ็จๆผๆดๅ ท้ซ็ๆณๅพไปปๅ๏ผๅฆ่ชๅๅคๆฑบๆธๅๆใๆณๅพๅฏฆ้ซ่ญๅฅ๏ผNER๏ผใๆณ่ฆ็ทจ่่ฝๆ๏ผไปฅๅๆณๅพๅ่ฆๅฏฉๆฅ่ผๅฉใๆญคๆจกๅๅฏไปฅ็ก็ธซ้ๆ่ณๆณๅพๆธๆ็งๅญธๆ็จๆๆณๅพๆ่ก๏ผLegalTech๏ผ็ณป็ตฑไธญ๏ผๅนซๅฉๆณๅพๅฐๆฅญไบบๅฃซๆไผๆฅญๆๅๅทฅไฝๆ็ใ
Out-of-Scope Use
่ฉฒๆจกๅไธฆไธ้ฉ็จๆผ้ๆณๅพ็ธ้้ ๅ็็ๆไปปๅ๏ผไธไธๆ็จๆผ้ฒ่กๅฏ่ฝๆถๅ่ชคๅฐๆ้ฏ่ชค็ๆณๅพๅปบ่ญฐ๏ผๅฐคๅ ถๆฏๅจๆช็ถๅฐๆฅญๅฏฉๆฅ็ๆ ๆณไธใ้ฟๅ ๅฐๆจกๅ็จๆผๆช็ถๆๆฌๆ้ๆณ็จ้๏ผๅฆ็ๆๅ ท็ญ่ญฐๆงๆๅ ทๅ่ฆ็ๆณๅพๅปบ่ญฐใ
Bias, Risks, and Limitations
ๆจกๅๅจ็ๆๆณๅพๆขๆๅๅคๆฑบๆธๅ งๅฎนๆ๏ผๅฏ่ฝๆ็ๆ่ๆงๆไธๅญๅจ็ๆณๆขๆๅคๆฑบๆธๅ งๅฎน๏ผ้ๆฏๆจกๅ็ๅ งๅจ้ๅถไนไธใไฝฟ็จ่ ๅจๅ่้ไบ่ณๆๆ๏ผๆ่ฌนๆ ๆชขๆฅ็ๆ็ๅ งๅฎน๏ผไธฆ้ฟๅ ๅฐๆจกๅ่ผธๅบ่ฆ็บๆณๅพไพๆใๅปบ่ญฐๅจๅฏฆ้ๆ็จไธญ๏ผๅฐๆจกๅ็ๆ็็ตๆ่ๅฏ้ ็ๆณๅพ่ฆ่งฃๅไพๆบ้ฒ่กๆฏๅฐ๏ผ็ขบไฟๆบ็ขบๆงใๅๆณๆงๅ้ฉ็จๆงใ
Recommendations
ๆญคๆจกๅ้็ถ็ถ้ๆณๅพๆๆฌ็ๅพฎ่ชฟ๏ผไฝๅจๆผๆณๅพๆๆฌ็ๆธ้ๅๅบ็คๆจกๅ็บ SLM๏ผๆจกๅ่ฝๅไปๆๆฅต้๏ผไฝฟ็จ่ ๆๆณจๆไปฅไธ้ขจ้ช่้ๅถ๏ผ
ๅ่ฆ้ขจ้ช๏ผ ๆจกๅๅฏ่ฝๆๅๆ ๅ ถ่จ็ทด่ณๆไธญ็ๆฝๅจๅ่ฆใ็ฑๆผๆณๅพๆๆฌ็็นๅฎๆง๏ผๆจกๅๅฏ่ฝๆด็ๆๆไบๆณ่ฆใๆขๆๆๅคๆฑบๆกไพ๏ผ่ๅจๅ ถไป้ ๅ่กจ็พ่ผๅผฑใ็นๅฅๆฏๅจ่็ไธๅธธ่ฆ็ๆณๅพๅ้กๆๆช่ขซ่จ็ทด้็ๆฐๆณ่ฆๆ๏ผๆจกๅ็่ผธๅบๅฏ่ฝๆๅธถๆๅ่ฆใ
ๆ่ก้ๅถ๏ผ ้็ถๆจกๅ่ฝๅค ่็ๅคง้จๅ็ๆณๅพๆๆฌ๏ผไฝๅฐๆผ็ตๆงๆฅตๅ ถ่ค้ๆ่ช่จๆจกๆฃฑๅ ฉๅฏ็ๆณๅพๆขๆ๏ผๆจกๅๅฏ่ฝ็กๆณ็ข็็ฒพ็ขบ็ๅ็ญใไฝฟ็จ่ ๆ้ฟๅ ๅฎๅ จไพ่ณดๆจกๅ็่ผธๅบ๏ผๅฐคๅ ถๅจๆณๅพๆฑบ็ญ้็จไธญ๏ผๅปบ่ญฐ้ฒ่ก้กๅค็ๅฐๆฅญๆชขๆฅใ
ๆณๅพ่ฒฌไปป๏ผ ๆจกๅไธฆ้ๅฐๆฅญๆณๅพ้กงๅ๏ผๅ ๆญคๅ ถ็ๆ็ๅๆไธๆ่ขซ่ฆ็บๆญฃ็ขบ็ๆณๅพๅปบ่ญฐใไฝฟ็จ่ ๆ็ขบไฟๅจ็ๆงไธๅฐๆฅญ่ๆฏไธ้ฒ่กๆจกๅ็ๆ็จ๏ผไธฆ้ฟๅ ๅจ้้ตๆฑบ็ญไธญ้ๅบฆไพ่ณดๆจกๅใ
่ชค็จ้ขจ้ช๏ผ ไธ็ถไฝฟ็จๆจกๅ้ฒ่ก้ฏ่ชคๆ่ชคๅฐๆง็ๆณๅพๅปบ่ญฐ๏ผๅฏ่ฝๅฐๅไบบๆไผๆฅญ้ ๆ่ฒ ้ขๅฝฑ้ฟใไฝฟ็จ่ ๆ่ฌนๆ ๆ็จๆจกๅๆผๅ่ฆๆๆณๅพ็ธ้ไปปๅไธญ๏ผไธฆไฟๆๅฐๅ ถ่ผธๅบ็ๆชข่ฆๅๆ กๆญฃใ
็บไบๆธๅฐ้ไบ้ขจ้ช๏ผๅปบ่ญฐไฝฟ็จ่ ๅจๆ็จๆจกๅ่ผธๅบๆ้ฒ่กไบๆฌกๆชขๆฅ๏ผ็นๅฅๆฏๅจๆถๅๆณๅพๆฑบ็ญ็ๆ ๅขไธญใๆฌๆจกๅ็พ้ๆฎต็บๆไพๆณๅพ็งๆ้ ๅ้ฒ่กๅคง่ช่จๆจกๅ็ ็ฉถ๏ผไธฆ้ๅไปฃๅฐๆฅญๆณๅพๅทฅไฝ่ ไนๅฐๆฅญๅปบ่ญฐใ
How to Get Started with the Model
Using vLLM
่ฆไฝฟ็จ vLLM Docker image ไพๅๅๆญคๆจกๅ๏ผๆจๅฏไปฅๆ็ งไปฅไธๆไฝ๏ผ
docker run --runtime nvidia --gpus all \
-v ~/.cache/huggingface:/root/.cache/huggingface \
--env "HUGGING_FACE_HUB_TOKEN=<secret>" \
-p 8000:8000 \
--ipc=host \
vllm/vllm-openai:latest \
--model lianghsun/Llama-3.2-Taiwan-Legal-3B-Instruct
่ซๆณจๆ๏ผๅฆๆๆณ่ฆไฝฟ็จไธๅ็ๆฌ็ checkpoint๏ผ่ซๅ ไธ --revision <tag_name>
๏ผ
docker run --runtime nvidia --gpus all \
-v ~/.cache/huggingface:/root/.cache/huggingface \
--env "HUGGING_FACE_HUB_TOKEN=<secret>" \
-p 8000:8000 \
--ipc=host \
vllm/vllm-openai:latest \
--model lianghsun/Llama-3.2-Taiwan-Legal-3B-Instruct --revision <tag_name>
Training Details
Training Data (for v2024.11.13)
- lianghsun/tw-legal-nlp
- lianghsun/tw-legal-synthetic-qa
- lianghsun/tw-law-article-qa
- lianghsun/tw-judgment-qa
- lianghsun/tw-bar-examination-2020-chat
- lianghsun/tw-emergency-medicine-bench
- lianghsun/tw-structured-law-article
- lianghsun/tw-judgment-gist-chat
- lianghsun/vulnerability-mitigation-qa-zh_tw
- lianghsun/tw-legal-qa-chat
- lianghsun/reasoning-base-20k-chat
- lianghsun/tw-contract-review-chat
- rombodawg/Everything_Instruct_Multilingual
- benchang1110/Belle-Taide
Training procedure
Preprocessing
ๆญคๆจกๅๅจ v2024.11.06 ็ๆฌๅพๆนๆก็จ lianghsun/Llama-3.2-Taiwan-3B ๅ็บๅบ็คๆจกๅ๏ผfoundation model๏ผใ Tokenizer ไป่ๆๅๅ ็ meta-llama/Llama-3.2-3B ็ธๅ๏ผๅจๆชไพ็ๆฌๅ่ฆๆณๆดๅ ไธญๆๅญใ
Training hyperparameters (for v2024.11.13)
The following hyperparameters were used during training:
- learning_rate: (initial lr) 5e-5
- train_batch_size: 20
- eval_batch_size: Not specified
- seed: 42
- distributed_type: single-node
- num_devices: 8
- gradient_accumulation_steps: 16
- total_train_batch_size: 1,280 (train_batch_size * gradient_accumulation_steps * num_devices)
- optimizer: adamw_torch_fused
- lr_scheduler_type: cosine
- lr_scheduler_warmup_steps: 100
- num_epochs: 3
- grad_norm: 1.1764454343711086
- global_step: 65
Speeds, Sizes, Times (for v2024.11.13)
- Duration: 30m 19s
- Train runtime: 30m 19s
- Train samples per second: 1.1764454343711086
- Train steps per second: 0.036
- Total training FLOPs: 89,423,735,685,120
- Train loss: 0.7657
Evaluation
Testing Data, Factors & Metrics
Testing Data
Note: ..(WIP)..
Factors
Note: ..(WIP)..
Metrics
Note: ..(WIP)..
Results
Note: ..(WIP)..
Summary
Note: ..(WIP)..
Model Examination
ๆณๆขๅ่ฆ
Note: ..(WIP)..
ๅคๆฑบๆธๅ งๅฎน
Note: ..(WIP)..
ๆณๅพ NLP ไปปๅ
Note: ..(WIP)..
Environmental Impact (for v2024.11.13)
- Hardware Type: 8 x NVIDIA A100 40GB
- Hours used: 30m 19s
- Cloud Provider: N/A
- Compute Region: N/A
- Carbon Emitted: N/A
Technical Specifications
Model Architecture and Objective
ๆฌๆจกๅๅบๆผ lianghsun/Llama-3.2-Taiwan-3B ๏ผไฝฟ็จ่ชๅๆญธ Transformer ๆถๆง้ฒ่ก่ช่จๅปบๆจกใ่ฉฒๆจกๅ็ไธป่ฆ็ฎๆจๆฏๆๅๅฐๅฐ็ฃๆณๅพๆๆฌ็็่งฃ่็ๆ่ฝๅ๏ผๅฐคๅ ถๆฏ้ๅฐๅคๆฑบๆธใๆณๆข็ๅฐๆฅญ่็่ๆ็จใ้้ๅฐ้่จญ่จ็ๆณๅพๆๆฌ้้ฒ่กๅพฎ่ชฟ๏ผๆจกๅ่ฝๆด็ฒพ็ขบๅฐๅ็ญๆณๅพๅ้กไธฆๆไพ็ธ้ๅปบ่ญฐใ
Compute Infrastructure
Hardware (for v2024.11.6)
- 8 x NVIDIA A100 40GB
Software
- ๅพฎ่ชฟ้็จไฝฟ็จไบ hiyouga/LLaMA-Factory ๆกๆถ้ฒ่ก่จ็ทดใ
Citation
็กใ
Glossary
็กใ
More Information
็ฎๅ
ๅ็ฎกๆๅๅทฒๆบๅไบ่จฑๅค้ๆผไธญ่ฏๆฐๅๅฐ็ฃๆณๅพ้ ๅ็่ณๆ้๏ผไฝ็ฑๆผ็ฎๅ่ณๆบๆ้๏ผ็กๆณๅฐๆๆ่ณๆ้้ฒ่กๅฎๆด่จ็ทด๏ผๆฏ็๏ผๆๅไธฆๆฒๆๅฐๅ จ้จ่ณๆ้้ฝ้ฒ่ก่จ็ทด๏ผๅ ๅๅบ่ขซ่ช็บๆๅบ็ค็ๆณๅพๆๆฌ๏ผ๏ผๅฐ่ดๆจกๅๅฐๆช้ๅฐๆไฝณ่กจ็พใๅ ๆญค๏ผ็ฎๅ็ checkpoint ๆฏๅบๆผๆ้่ณๆบ็็ๆฌใๅฆๆๆจๆๆ้ก่ดๅฉ็ฎๅ๏ผๆญก่ฟ่ๆ่ฏ็นซใๆ็ธไฟก่ฅ่ฝๅฐๆดๅคๅทฒๆบๅไฝๅฐๆช็ดๅ ฅ่จ็ทด็ๆณๅพ่ชๆ้ฒ่กๅพฎ่ชฟ๏ผ่ฉฒๆจกๅๅฐ่ฝ้ๅฐ็น้ซไธญๆๆณๅพ้ ๅ็ๆไฝณ่กจ็พใ
ๆ็ธพๆดๆฐ
ๆญคๆจกๅๅฆๆ้ฒไธๆญฅ่ณๆบ๏ผๅฐๆไธๅฎๆๆดๆฐ๏ผๆ้ๆจกๅๆๆฐๆถๆฏ่ซ่ฆ Model Update History ็ซ ็ฏใ
Model Card Authors
Model Card Contact
Framework versions
- Transformers 4.45.2
- Pytorch 2.4.1+cu121
- Datasets 2.21.0
- Tokenizers 0.20.0
- Downloads last month
- 5
Model tree for lianghsun/Llama-3.2-Taiwan-Legal-3B-Instruct
Datasets used to train lianghsun/Llama-3.2-Taiwan-Legal-3B-Instruct
Spaces using lianghsun/Llama-3.2-Taiwan-Legal-3B-Instruct 8
Collection including lianghsun/Llama-3.2-Taiwan-Legal-3B-Instruct
Evaluation results
- single choice on tw-legal-benchmark-v1self-reported22.010