Developed by :

  • Changgil Song

Model Number:

  • k2s3_test_24001

Base Model :

Training Data

  • The model was trained on a diverse dataset comprising approximately 800 million tokens, including the Standard Korean Dictionary, KULLM training data from Korea University, dissertation abstracts from master's and doctoral theses, and Korean language samples from AI Hub.
  • 이 λͺ¨λΈμ€ ν‘œμ€€λŒ€κ΅­μ–΄μ‚¬μ „, κ³ λ €λŒ€ KULLM의 ν›ˆλ ¨ 데이터, μ„λ°•μ‚¬ν•™μœ„μž μ„œμ§€μ •λ³΄ λ…Όλ¬Έμ΄ˆλ‘, ai_hub의 ν•œκ΅­μ–΄ 데이터 μƒ˜ν”Œλ“€μ„ ν¬ν•¨ν•˜μ—¬ μ•½ 8μ–΅ 개의 ν† ν°μœΌλ‘œ κ΅¬μ„±λœ λ‹€μ–‘ν•œ λ°μ΄ν„°μ…‹μ—μ„œ ν›ˆλ ¨λ˜μ—ˆμŠ΅λ‹ˆλ‹€.

Training Method

  • This model was fine-tuned on the "meta-llama/Llama-2-13b-chat-hf" base model using PEFT (Parameter-Efficient Fine-Tuning) LoRA (Low-Rank Adaptation) techniques.
  • 이 λͺ¨λΈμ€ "meta-llama/Llama-2-13b-chat-hf" 기반 λͺ¨λΈμ„ PEFT LoRAλ₯Ό μ‚¬μš©ν•˜μ—¬ λ―Έμ„Έμ‘°μ •λ˜μ—ˆμŠ΅λ‹ˆλ‹€.

Hardware and Software

  • Hardware: Utilized two A100 (80G*2EA) GPUs for training.
  • Training Factors: This model was fine-tuned using PEFT LoRA with the HuggingFace SFTtrainer and applied fsdp. Key parameters included LoRA r = 8, LoRA alpha = 16, trained for 2 epochs, batch size of 1, and gradient accumulation of 32.
  • 이 λͺ¨λΈμ€ PEFT LoRAλ₯Ό μ‚¬μš©ν•˜μ—¬ HuggingFace SFTtrainer와 fsdpλ₯Ό μ μš©ν•˜μ—¬ λ―Έμ„Έμ‘°μ •λ˜μ—ˆμŠ΅λ‹ˆλ‹€. μ£Όμš” νŒŒλΌλ―Έν„°λ‘œλŠ” LoRA r = 8, LoRA alpha = 16, 2 에폭 ν›ˆλ ¨, 배치 크기 1, 그리고 κ·ΈλΌλ””μ–ΈνŠΈ λˆ„μ  32λ₯Ό ν¬ν•¨ν•©λ‹ˆλ‹€.

Caution

  • For fine-tuning this model, it is advised to consider the specific parameters used during training, such as LoRA r and LoRA alpha values, to ensure compatibility and optimal performance.
  • 이 λͺ¨λΈμ„ λ―Έμ„Έμ‘°μ •ν•  λ•ŒλŠ” LoRA r 및 LoRA alpha κ°’κ³Ό 같이 ν›ˆλ ¨ 쀑에 μ‚¬μš©λœ νŠΉμ • νŒŒλΌλ―Έν„°λ₯Ό κ³ λ €ν•˜λŠ” 것이 μ’‹μŠ΅λ‹ˆλ‹€. μ΄λŠ” ν˜Έν™˜μ„± 및 졜적의 μ„±λŠ₯을 보μž₯ν•˜κΈ° μœ„ν•¨μž…λ‹ˆλ‹€.

Additional Information

  • The training leveraged the fsdp (Fully Sharded Data Parallel) feature through the HuggingFace SFTtrainer for efficient memory usage and accelerated training.
  • ν›ˆλ ¨μ€ HuggingFace SFTtrainerλ₯Ό ν†΅ν•œ fsdp κΈ°λŠ₯을 ν™œμš©ν•˜μ—¬ λ©”λͺ¨λ¦¬ μ‚¬μš©μ„ 효율적으둜 ν•˜κ³  ν›ˆλ ¨ 속도λ₯Ό κ°€μ†ν™”ν–ˆμŠ΅λ‹ˆλ‹€.
Downloads last month
86
Inference Providers NEW
Input a message to start chatting with Changgil/k2s3_test_24001.

Model tree for Changgil/k2s3_test_24001

Quantizations
1 model

Spaces using Changgil/k2s3_test_24001 9