Edit model card

Model Card for RedWhale-tv-10.8B-ipt-v0.1

Model Description

The RedWhale-tv-10.8B-ipt-v0.1 is an Instruction Pre-Trained (IPT) version of the RedWhale-tv-10.8B-v1.0, created through continual training for 5000 steps using 80,000 single-turn synthetic instruction data points (not multi-turn). The training was performed on a single NVIDIA A5000 24GB GPU using the Low-Rank Adaptation (LoRA) method.

Multi-turn instruction data will be explored in future iterations.

The model μ‚¬μš©μ„ μ›ν•˜μ‹œλ©΄ repo access μš”μ²­ν•΄μ£Όμ„Έμš”.

About the Model

Load the Model

from transformers import AutoTokenizer
from transformers import AutoModelForCausalLM

YOUR_HF_TOKEN_READ = "hf_..."

model_name_or_path = "TwinDoc/RedWhale-tv-10.8B-ipt-v0.1"

tokenizer = AutoTokenizer.from_pretrained(model_name_or_path, token=YOUR_HF_TOKEN_READ)
model = AutoModelForCausalLM.from_pretrained(model_name_or_path, token=YOUR_HF_TOKEN_READ)

Generate Text

messages = [
  {'content': '당신은 λ‹€μ–‘ν•œ μž‘μ—…μ— λŒ€ν•œ ν•œκ΅­μ–΄ 지침을 μ œκ³΅ν•˜λ„λ‘ ν›ˆλ ¨λœ λ‹€κ΅­μ–΄ AI λͺ¨λΈμž…λ‹ˆλ‹€.', 'role': 'system'},
  {'content': 'ν•œκ΅­μ˜ 전톡 μŒμ‹μ€ λ¬΄μ—‡μΈκ°€μš”?', 'role': 'user'}
]

text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True, return_tensors="pt")
# text = '<s> [INST] 당신은 λ‹€μ–‘ν•œ μž‘μ—…μ— λŒ€ν•œ ν•œκ΅­μ–΄ 지침을 μ œκ³΅ν•˜λ„λ‘ ν›ˆλ ¨λœ λ‹€κ΅­μ–΄ AI λͺ¨λΈμž…λ‹ˆλ‹€.\n\nν•œκ΅­μ˜ 전톡 μŒμ‹μ€ λ¬΄μ—‡μΈκ°€μš”? [/INST]'

encodings = tokenizer(text, return_tensors='pt')
terminators = [tokenizer.eos_token_id] 
max_new_tokens = 64

outputs = model.generate(**encodings, eos_token_id=terminators, max_new_tokens=max_new_tokens)
generated_text = tokenizer.batch_decode(outputs)[0]
# generated_text = '<s>  [INST] 당신은 λ‹€μ–‘ν•œ μž‘μ—…μ— λŒ€ν•œ ν•œκ΅­μ–΄ 지침을 μ œκ³΅ν•˜λ„λ‘ ν›ˆλ ¨λœ λ‹€κ΅­μ–΄ AI λͺ¨λΈμž…λ‹ˆλ‹€.\n\nν•œκ΅­μ˜ 전톡 μŒμ‹μ€ λ¬΄μ—‡μΈκ°€μš”? [/INST] ν•œκ΅­μ˜ 전톡 μŒμ‹μ€ λ‹€μ–‘ν•œ 지역과 κ³„μ ˆμ— 따라 λ‹€μ–‘ν•œ μ’…λ₯˜κ°€ μžˆμŠ΅λ‹ˆλ‹€. λŒ€ν‘œμ μΈ 전톡 μŒμ‹μ€ λ‹€μŒκ³Ό κ°™μŠ΅λ‹ˆλ‹€.\n\n1. **λΉ„λΉ”λ°₯**: λΉ„λΉ”λ°₯은 λ‹€μ–‘ν•œ 재료λ₯Ό μ„žμ–΄ λ§Œλ“  λ°₯ μœ„μ— 양념을 뿌렀 λ¨ΉλŠ” μŒμ‹μž…λ‹ˆλ‹€.\n2. **κΉ€μΉ˜**: κΉ€μΉ˜λŠ” ν•œκ΅­μ˜ λŒ€ν‘œμ μΈ 발효 μ‹ν’ˆ'

Generate Streaming Text

from transformers import TextStreamer
text_streamer = TextStreamer(tokenizer)

messages = [
  {'content': '당신은 λ‹€μ–‘ν•œ μž‘μ—…μ— λŒ€ν•œ ν•œκ΅­μ–΄ 지침을 μ œκ³΅ν•˜λ„λ‘ ν›ˆλ ¨λœ λ‹€κ΅­μ–΄ AI λͺ¨λΈμž…λ‹ˆλ‹€.', 'role': 'system'},
  {'content': 'ν•œκ΅­μ˜ 전톡 μŒμ‹μ€ λ¬΄μ—‡μΈκ°€μš”?', 'role': 'user'}
]

text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True, return_tensors="pt")
# text = '<s> [INST] 당신은 λ‹€μ–‘ν•œ μž‘μ—…μ— λŒ€ν•œ ν•œκ΅­μ–΄ 지침을 μ œκ³΅ν•˜λ„λ‘ ν›ˆλ ¨λœ λ‹€κ΅­μ–΄ AI λͺ¨λΈμž…λ‹ˆλ‹€.\n\nν•œκ΅­μ˜ 전톡 μŒμ‹μ€ λ¬΄μ—‡μΈκ°€μš”? [/INST]'

encodings = tokenizer(text, return_tensors='pt')
terminators = [tokenizer.eos_token_id] 
max_new_tokens = 64

outputs = model.generate(**encodings, eos_token_id=terminators, max_new_tokens=max_new_tokens)
generated_text = model.generate(**encodings, streamer = text_streamer, max_new_tokens = max_new_tokens)
# generated_text = '<s>  [INST] 당신은 λ‹€μ–‘ν•œ μž‘μ—…μ— λŒ€ν•œ ν•œκ΅­μ–΄ 지침을 μ œκ³΅ν•˜λ„λ‘ ν›ˆλ ¨λœ λ‹€κ΅­μ–΄ AI λͺ¨λΈμž…λ‹ˆλ‹€.\n\nν•œκ΅­μ˜ 전톡 μŒμ‹μ€ λ¬΄μ—‡μΈκ°€μš”? [/INST] ν•œκ΅­μ˜ 전톡 μŒμ‹μ€ λ‹€μ–‘ν•œ 지역과 κ³„μ ˆμ— 따라 λ‹€μ–‘ν•œ μ’…λ₯˜κ°€ μžˆμŠ΅λ‹ˆλ‹€. λŒ€ν‘œμ μΈ 전톡 μŒμ‹μ€ λ‹€μŒκ³Ό κ°™μŠ΅λ‹ˆλ‹€.\n\n1. **λΉ„λΉ”λ°₯**: λΉ„λΉ”λ°₯은 λ‹€μ–‘ν•œ 재료λ₯Ό μ„žμ–΄ λ§Œλ“  λ°₯ μœ„μ— 양념을 뿌렀 λ¨ΉλŠ” μŒμ‹μž…λ‹ˆλ‹€.\n2. **κΉ€μΉ˜**: κΉ€μΉ˜λŠ” ν•œκ΅­μ˜ λŒ€ν‘œμ μΈ 발효 μ‹ν’ˆ'

License

The content of this project, created by AGILESODA, is licensed under the Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0).

Citation

@misc{vo2024redwhaleadaptedkoreanllm,
      title={RedWhale: An Adapted Korean LLM Through Efficient Continual Pretraining}, 
      author={Anh-Dung Vo and Minseong Jung and Wonbeen Lee and Daewoo Choi},
      year={2024},
      eprint={2408.11294},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2408.11294}, 
}

Built with:

AgileSoda TwinDoc Icon
Downloads last month
58
Safetensors
Model size
10.8B params
Tensor type
F32
Β·
Inference Examples
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.

Model tree for TwinDoc/RedWhale-tv-10.8B-ipt-v0.1

Finetuned
(1)
this model
Quantizations
2 models

Collection including TwinDoc/RedWhale-tv-10.8B-ipt-v0.1