Edit model card

AKALLAMA

AkaLlama is a series of Korean language models designed for practical usability across a wide range of tasks. The initial model, AkaLlama-v0.1, is a fine-tuned version of Meta-Llama-3-70b-Instruct. It has been trained on a custom mix of publicly available datasets curated by the MIR Lab. Our goal is to explore cost-effective ways to adapt high-performing LLMs for specific use cases, such as different languages (e.g., Korean) or domains (e.g., organization-specific chatbots).

For details, check out our project page.

Model Description

This is the model card of a GGUF model that has been pushed on the Hub.

About GGUF

GGUF is a new format introduced by the llama.cpp team on August 21st 2023. It is a replacement for GGML, which is no longer supported by llama.cpp.

Here is an incomplete list of clients and libraries that are known to support GGUF:

  • llama.cpp. The source project for GGUF. Offers a CLI and a server option.
  • text-generation-webui, the most widely used web UI, with many features and powerful extensions. Supports GPU acceleration.
  • KoboldCpp, a fully featured web UI, with GPU accel across all platforms and GPU architectures. Especially good for story telling.
  • GPT4All, a free and open source local running GUI, supporting Windows, Linux and macOS with full GPU accel.
  • LM Studio, an easy-to-use and powerful local GUI for Windows and macOS (Silicon), with GPU acceleration. Linux available, in beta as of 27/11/2023.
  • LoLLMS Web UI, a great web UI with many interesting and unique features, including a full model library for easy model selection.
  • Faraday.dev, an attractive and easy to use character-based chat GUI for Windows and macOS (both Silicon and Intel), with GPU acceleration.
  • llama-cpp-python, a Python library with GPU accel, LangChain support, and OpenAI-compatible API server.
  • candle, a Rust ML framework with a focus on performance, including GPU support, and ease of use.
  • ctransformers, a Python library with GPU accel, LangChain support, and OpenAI-compatible AI server. Note, as of time of writing (November 27th 2023), ctransformers has not been updated in a long time and does not support many recent models.

How to use

This repo provides gguf weight files for AkaLlama-70B-v0.1.

Use with llama.cpp.python

See the snippet below for usage with llama.cpp.python:

from llama_cpp import Llama

# Set gpu_layers to the number of layers to offload to GPU. Set to 0 if no GPU acceleration is available on your system.
llm = Llama(
  model_path="./AkaLlama-llama3-70b-v0.1.Q4_K_M.gguf",  # Download the model file first
  n_ctx=8192,  # The max sequence length to use - note that longer sequence lengths require much more resources
  n_threads=8,            # The number of CPU threads to use, tailor to your system and the resulting performance
  n_gpu_layers=81         # The number of layers to offload to GPU, if you have GPU acceleration available
)

# Simple inference example
output = llm(
  """<|begin_of_text|><|start_header_id|>system<|end_header_id|>

๋‹น์‹ ์€ ์—ฐ์„ธ๋Œ€ํ•™๊ต ๋ฉ€ํ‹ฐ๋ชจ๋‹ฌ ์—ฐ๊ตฌ์‹ค (MIR lab) ์ด ๋งŒ๋“  ๋Œ€๊ทœ๋ชจ ์–ธ์–ด ๋ชจ๋ธ์ธ AkaLlama (์•„์นด๋ผ๋งˆ) ์ž…๋‹ˆ๋‹ค.
๋‹ค์Œ ์ง€์นจ์„ ๋”ฐ๋ฅด์„ธ์š”:
1. ์‚ฌ์šฉ์ž๊ฐ€ ๋ณ„๋„๋กœ ์š”์ฒญํ•˜์ง€ ์•Š๋Š” ํ•œ ํ•ญ์ƒ ํ•œ๊ธ€๋กœ ์†Œํ†ตํ•˜์„ธ์š”.
2. ์œ ํ•ดํ•˜๊ฑฐ๋‚˜ ๋น„์œค๋ฆฌ์ , ์ฐจ๋ณ„์ , ์œ„ํ—˜ํ•˜๊ฑฐ๋‚˜ ๋ถˆ๋ฒ•์ ์ธ ๋‚ด์šฉ์ด ๋‹ต๋ณ€์— ํฌํ•จ๋˜์–ด์„œ๋Š” ์•ˆ ๋ฉ๋‹ˆ๋‹ค.
3. ์งˆ๋ฌธ์ด ๋ง์ด ๋˜์ง€ ์•Š๊ฑฐ๋‚˜ ์‚ฌ์‹ค์— ๋ถ€ํ•ฉํ•˜์ง€ ์•Š๋Š” ๊ฒฝ์šฐ ์ •๋‹ต ๋Œ€์‹  ๊ทธ ์ด์œ ๋ฅผ ์„ค๋ช…ํ•˜์„ธ์š”. ์งˆ๋ฌธ์— ๋Œ€ํ•œ ๋‹ต์„ ๋ชจ๋ฅธ๋‹ค๋ฉด ๊ฑฐ์ง“ ์ •๋ณด๋ฅผ ๊ณต์œ ํ•˜์ง€ ๋งˆ์„ธ์š”.
4. ์•ˆ์ „์ด๋‚˜ ์œค๋ฆฌ์— ์œ„๋ฐฐ๋˜์ง€ ์•Š๋Š” ํ•œ ์‚ฌ์šฉ์ž์˜ ๋ชจ๋“  ์งˆ๋ฌธ์— ์™„์ „ํ•˜๊ณ  ํฌ๊ด„์ ์œผ๋กœ ๋‹ต๋ณ€ํ•˜์„ธ์š”.<|eot_id|><|start_header_id|>user<|end_header_id|>

{prompt}<|eot_id|><|start_header_id|>assistant<|end_header_id|>

""", # Prompt
  max_tokens=512,  # Generate up to 512 tokens
  stop=["<|eot_id|>", "<|end_of_text|>"],   # Example stop token - not necessarily correct for this specific model! Please check before using.
  echo=True        # Whether to echo the prompt
)

# Chat Completion API

llm = Llama(model_path="./AkaLlama-llama3-70b-v0.1.Q4_K_M.gguf", chat_format="llama-3")  # Set chat_format according to the model you are using
llm.create_chat_completion(
    messages = [
        {"role": "system", "content": """๋‹น์‹ ์€ ์—ฐ์„ธ๋Œ€ํ•™๊ต ๋ฉ€ํ‹ฐ๋ชจ๋‹ฌ ์—ฐ๊ตฌ์‹ค (MIR lab) ์ด ๋งŒ๋“  ๋Œ€๊ทœ๋ชจ ์–ธ์–ด ๋ชจ๋ธ์ธ AkaLlama (์•„์นด๋ผ๋งˆ) ์ž…๋‹ˆ๋‹ค.
๋‹ค์Œ ์ง€์นจ์„ ๋”ฐ๋ฅด์„ธ์š”:
1. ์‚ฌ์šฉ์ž๊ฐ€ ๋ณ„๋„๋กœ ์š”์ฒญํ•˜์ง€ ์•Š๋Š” ํ•œ ํ•ญ์ƒ ํ•œ๊ธ€๋กœ ์†Œํ†ตํ•˜์„ธ์š”.
2. ์œ ํ•ดํ•˜๊ฑฐ๋‚˜ ๋น„์œค๋ฆฌ์ , ์ฐจ๋ณ„์ , ์œ„ํ—˜ํ•˜๊ฑฐ๋‚˜ ๋ถˆ๋ฒ•์ ์ธ ๋‚ด์šฉ์ด ๋‹ต๋ณ€์— ํฌํ•จ๋˜์–ด์„œ๋Š” ์•ˆ ๋ฉ๋‹ˆ๋‹ค.
3. ์งˆ๋ฌธ์ด ๋ง์ด ๋˜์ง€ ์•Š๊ฑฐ๋‚˜ ์‚ฌ์‹ค์— ๋ถ€ํ•ฉํ•˜์ง€ ์•Š๋Š” ๊ฒฝ์šฐ ์ •๋‹ต ๋Œ€์‹  ๊ทธ ์ด์œ ๋ฅผ ์„ค๋ช…ํ•˜์„ธ์š”. ์งˆ๋ฌธ์— ๋Œ€ํ•œ ๋‹ต์„ ๋ชจ๋ฅธ๋‹ค๋ฉด ๊ฑฐ์ง“ ์ •๋ณด๋ฅผ ๊ณต์œ ํ•˜์ง€ ๋งˆ์„ธ์š”.
4. ์•ˆ์ „์ด๋‚˜ ์œค๋ฆฌ์— ์œ„๋ฐฐ๋˜์ง€ ์•Š๋Š” ํ•œ ์‚ฌ์šฉ์ž์˜ ๋ชจ๋“  ์งˆ๋ฌธ์— ์™„์ „ํ•˜๊ณ  ํฌ๊ด„์ ์œผ๋กœ ๋‹ต๋ณ€ํ•˜์„ธ์š”."""},
        {
            "role": "user",
            "content": "๋„ค ์ด๋ฆ„์€ ๋ญ์•ผ?."
        }
    ]
)

# ๋‚ด ์ด๋ฆ„์€ AkaLlama์ž…๋‹ˆ๋‹ค! ๋‚˜๋Š” ์–ธ์–ด ๋ชจ๋ธ๋กœ, ์‚ฌ์šฉ์ž์™€ ๋Œ€ํ™”ํ•˜๋Š” ๋ฐ ๋„์›€์„ ์ฃผ๊ธฐ ์œ„ํ•ด ๋งŒ๋“ค์–ด์กŒ์Šต๋‹ˆ๋‹ค. ๋‚˜๋Š” ๋‹ค์–‘ํ•œ ์ฃผ์ œ์— ๋Œ€ํ•œ ์งˆ๋ฌธ์— ๋‹ตํ•˜๊ณ , ์ƒˆ๋กœ์šด ์•„์ด๋””์–ด๋ฅผ ์ œ๊ณตํ•˜๋ฉฐ, ๋ฌธ์ œ๋ฅผ ํ•ด๊ฒฐํ•˜๋Š” ๋ฐ ๋„์›€์ด ๋  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์‚ฌ์šฉ์ž๊ฐ€ ์›ํ•˜๋Š” ์ •๋ณด๋‚˜ ๋„์›€์„ ๋ฐ›๋„๋ก ์ตœ์„ ์„ ๋‹คํ•  ๊ฒƒ์ž…๋‹ˆ๋‹ค!

Compatibility

These quantised GGUFv2 files are compatible with llama.cpp from August 27th onwards, as of commit d0cee0d

They are also compatible with many third party UIs and libraries - please see the list at the top of this README.

Explanation of quantisation methods

Click to see details

The new methods available are:

  • GGML_TYPE_Q2_K - "type-1" 2-bit quantization in super-blocks containing 16 blocks, each block having 16 weight. Block scales and mins are quantized with 4 bits. This ends up effectively using 2.5625 bits per weight (bpw)
  • GGML_TYPE_Q3_K - "type-0" 3-bit quantization in super-blocks containing 16 blocks, each block having 16 weights. Scales are quantized with 6 bits. This end up using 3.4375 bpw.
  • GGML_TYPE_Q4_K - "type-1" 4-bit quantization in super-blocks containing 8 blocks, each block having 32 weights. Scales and mins are quantized with 6 bits. This ends up using 4.5 bpw.
  • GGML_TYPE_Q5_K - "type-1" 5-bit quantization. Same super-block structure as GGML_TYPE_Q4_K resulting in 5.5 bpw
  • GGML_TYPE_Q6_K - "type-0" 6-bit quantization. Super-blocks with 16 blocks, each block having 16 weights. Scales are quantized with 8 bits. This ends up using 6.5625 bpw

Refer to the Provided Files table below to see what files use which methods, and how.

Provided files

Name Quant method Bits Size Max RAM required Use case
AkaLlama-llama3-70b-v0.1.Q2_K.gguf Q2_K 2 26.4 GB 28.9 GB smallest, significant quality loss - not recommended for most purposes
AkaLlama-llama3-70b-v0.1.Q3_K_S.gguf Q3_K_S 3 30.9 GB 33.4 GB very small, high quality loss
AkaLlama-llama3-70b-v0.1.Q3_K_M.gguf Q3_K_M 3 34.3 GB 36.8 GB very small, high quality loss
AkaLlama-llama3-70b-v0.1.Q3_K_L.gguf Q3_K_L 3 37.1 GB 39.6 GB small, substantial quality loss
AkaLlama-llama3-70b-v0.1.Q4_K_S.gguf Q4_K_S 4 40.3 GB 42.8 GB small, greater quality loss
AkaLlama-llama3-70b-v0.1.Q4_K_M.gguf Q4_K_M 4 42.5 GB 45.0 GB medium, balanced quality - recommended
AkaLlama-llama3-70b-v0.1.Q5_K_S.gguf Q5_K_S 5 48.7 GB 50.2 GB large, low quality loss - recommended
AkaLlama-llama3-70b-v0.1.Q5_K_M.gguf Q5_K_M 5 50.0 GB 52.5 GB large, very low quality loss - recommended
AkaLlama-llama3-70b-v0.1.Q6_K.gguf Q6_K 6 54.4 GB 59.9 GB very large, extremely low quality loss
AkaLlama-llama3-70b-v0.1.Q8_0.gguf Q8_0 8 70.0 GB 72.5 GB very large, extremely low quality loss - not recommended

Note: the above RAM figures assume no GPU offloading. If layers are offloaded to the GPU, this will reduce RAM usage and use VRAM instead.

Q6_K and Q8_0 files are split and require joining

Note: HF does not support uploading files larger than 50GB. Therefore I have uploaded the Q6_K and Q8_0 files as split files.

q6_K

Please download:

  • AkaLlama-llama3-70b-v0.1.Q6_K.00001-of-00002.gguf
  • AkaLlama-llama3-70b-v0.1.Q6_K.00002-of-00002.gguf

q8_0

Please download:

  • AkaLlama-llama3-70b-v0.1.Q8_0.00001-of-00002.gguf
  • AkaLlama-llama3-70b-v0.1.Q8_0.00002-of-00002.gguf

To join the files, do the following:

Linux and macOS:

cat AkaLlama-llama3-70b-v0.1.Q6_K.*-of-00002.gguf > AkaLlama-llama3-70b-v0.1.Q6_K.gguf && rm AkaLlama-llama3-70b-v0.1.Q6_K.*-of-00002.gguf
cat AkaLlama-llama3-70b-v0.1.Q8_0.*-of-00002.gguf > AkaLlama-llama3-70b-v0.1.Q8_0.gguf && rm AkaLlama-llama3-70b-v0.1.Q8_0.*-of-00002.gguf

Windows command line:

COPY /B AkaLlama-llama3-70b-v0.1.Q6_K.00001-of-00002.gguf + AkaLlama-llama3-70b-v0.1.Q6_K.00002-of-00002.gguf AkaLlama-llama3-70b-v0.1.Q6_K.gguf
del AkaLlama-llama3-70b-v0.1.Q6_K.00001-of-00002.gguf AkaLlama-llama3-70b-v0.1.Q6_K.00002-of-00002.gguf

COPY /B AkaLlama-llama3-70b-v0.1.Q8_0.00001-of-00002.gguf + AkaLlama-llama3-70b-v0.1.Q8_0.00002-of-00002.gguf AkaLlama-llama3-70b-v0.1.Q8_0.gguf
del AkaLlama-llama3-70b-v0.1.Q8_0.00001-of-00002.gguf AkaLlama-llama3-70b-v0.1.Q8_0.00002-of-00002.gguf

Evaluation

Model #Parameter Qunatized? LogicKor*
AkaLlama-llama3-70b-v0.1-GGUF.Q4 70B 4bit 6.56
AkaLlama-llama3-70b-v0.1-GGUF.Q8 70B 8bit 6.34

*mean over 3 random seeds

Training Details

Training Procedure

We trained AkaLlama using a preference learning alignment algorithm called Odds Ratio Preference Optimization (ORPO). Our training pipeline is almost identical to that of HuggingFaceH4/zephyr-orpo-141b-A35b-v0.1, aside from minor hyperparameter changes. Please check out Huggingface's alignment handbook for further details, including the chat template.

Training Data

Detailed descriptions regarding training data will be announced later.

Examples

User:
  ๋„ˆ๋Š” ๋ˆ„๊ตฌ์ด๊ณ  ์–ด๋–ค ์ผ์„ ํ•  ์ˆ˜ ์žˆ์–ด?
Model:
  ์ €๋Š” ์ธ๊ณต์ง€๋Šฅ ์–ธ์–ด ๋ชจ๋ธ์ธ AkaLlama(์•„์นด๋ผ๋งˆ)์ž…๋‹ˆ๋‹ค. ์ €๋Š” ๋‹ค์–‘ํ•œ ์ฃผ์ œ์™€ ํ˜•์‹์˜ ํ…์ŠคํŠธ๋ฅผ ์ƒ์„ฑํ•  ์ˆ˜ ์žˆ์œผ๋ฉฐ, ์งˆ๋ฌธ์— ๋‹ตํ•˜๊ณ , ๋Œ€ํ™”๋ฅผ ์ง„ํ–‰ํ•˜๋ฉฐ, ์ด์•ผ๊ธฐ๋ฅผ ๋งŒ๋“ค ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๋˜ํ•œ ์‚ฌ์šฉ์ž๊ฐ€ ์ง€์ •ํ•œ ์ฃผ์ œ๋‚˜ ์žฅ๋ฅด์— ๋”ฐ๋ผ ์ฐฝ์˜์ ์ธ ๊ธ€์“ฐ๊ธฐ๋‚˜ ํŽธ์ง‘ ์ž‘์—…์„ ์ˆ˜ํ–‰ํ•  ์ˆ˜๋„ ์žˆ์Šต๋‹ˆ๋‹ค.
Math Solving[CLICK TO EXPAND]
User:
  ์•„๋ž˜ ๋ฌธ์ œ๋ฅผ ๊นŠ๊ฒŒ ์ƒ๊ฐํ•˜๋ฉด์„œ ํ’€์–ด๋ด
  ์—ฐํ•„ 5 ํƒ€์™€ ์ง€์šฐ๊ฐœ 45 ๊ฐœ๊ฐ€ ์žˆ์Šต๋‹ˆ๋‹ค. ์ด๊ฒƒ์„ ๋  ์ˆ˜ ์žˆ๋Š” ๋Œ€๋กœ ๋งŽ์€ ํ•™์ƒ๋“ค์—๊ฒŒ ๋‚จ๊น€์—†์ด ๋˜‘๊ฐ™์ด ๋‚˜๋ˆ„์–ด ์ฃผ๋ ค๊ณ  ํ•ฉ๋‹ˆ๋‹ค. ๋ช‡ ๋ช…๊นŒ์ง€ ๋‚˜๋ˆ„์–ด ์ค„ ์ˆ˜ ์žˆ์Šต๋‹ˆ๊นŒ?
Model:
  ์•„๋ž˜ ๋ฌธ์ œ๋ฅผ ๊นŠ๊ฒŒ ์ƒ๊ฐํ•˜๋ฉด์„œ ํ’€์–ด๋ด…์‹œ๋‹ค:
  ์—ฐํ•„ 5 ํƒ€์™€ ์ง€์šฐ๊ฐœ 45 ๊ฐœ๊ฐ€ ์žˆ์Šต๋‹ˆ๋‹ค. ์ด๊ฒƒ์„ ๋  ์ˆ˜ ์žˆ๋Š” ๋Œ€๋กœ ๋งŽ์€ ํ•™์ƒ๋“ค์—๊ฒŒ ๋‚จ๊น€์—†์ด ๋˜‘๊ฐ™์ด ๋‚˜๋ˆ„์–ด ์ฃผ๋ ค๊ณ  ํ•ฉ๋‹ˆ๋‹ค. ๋ช‡ ๋ช…๊นŒ์ง€ ๋‚˜๋ˆ„์–ด ์ค„ ์ˆ˜ ์žˆ์Šต๋‹ˆ๊นŒ?
  ํ’€์ด:
  1. ์—ฐํ•„ 5 ํƒ€๋ฅผ 1 ๊ฐœ์”ฉ ๋‚˜๋ˆ ์„œ ๋ช‡ ๋ช…์—๊ฒŒ ๋‚˜๋ˆ„์–ด ์ค„ ์ˆ˜ ์žˆ๋Š”์ง€ ๋จผ์ € ๊ตฌํ•ฉ๋‹ˆ๋‹ค.
  - ๊ฐ ์—ฐํ•„์—๋Š” 12๊ฐœ์˜ ์—ฐํ•„์ด ํฌํ•จ๋˜์–ด ์žˆ์Šต๋‹ˆ๋‹ค. ๋”ฐ๋ผ์„œ ์ด 12 * 5 = 60๊ฐœ์ž…๋‹ˆ๋‹ค.
  - 60๊ฐœ๋ฅผ 1๋ช…๋‹น 1๊ฐœ์”ฉ ๋‚˜๋ˆ„๋ฉด 60๋ช…์˜ ํ•™์ƒ๋“ค์—๊ฒŒ ๋‚˜๋ˆ ์ค„ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
  2. ์ง€์šฐ๊ฐœ 45 ๊ฐœ๋ฅผ 1 ๊ฐœ์”ฉ ๋‚˜๋ˆ„์–ด ๋ช‡ ๋ช…์—๊ฒŒ ๋‚˜๋ˆ„์–ด ์ค„ ์ˆ˜ ์žˆ๋Š”์ง€ ๊ตฌํ•ฉ๋‹ˆ๋‹ค.
  - 45 ๊ฐœ๋ฅผ 1 ๋ช…๋‹น 1 ๊ฐœ์”ฉ ๋‚˜๋ˆ„๋ฉด 45 ๋ช…์˜ ํ•™์ƒ๋“ค์—๊ฒŒ ๋‚˜๋ˆ ์ค„ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
  3. ์—ฐํ•„๊ณผ ์ง€์šฐ๊ฐœ์˜ ๋‘ ๊ฐ€์ง€ ๋ฌผํ’ˆ์„ ํ•จ๊ป˜ ๊ณ ๋ คํ•˜์—ฌ ๊ฐ€์žฅ ๋งŽ์€ ํ•™์ƒ๋“ค์—๊ฒŒ ๋˜‘๊ฐ™์ด ๋‚˜๋ˆ„์–ด ์ค„ ์ˆ˜ ์žˆ๋Š” ๋ฐฉ๋ฒ•์„ ์ฐพ์Šต๋‹ˆ๋‹ค.
  - ์—ฐํ•„๋กœ๋งŒ 60๋ช…, ์ง€์šฐ๊ฐœ๋กœ๋งŒ 45๋ช…์˜ ํ•™์ƒ์—๊ฒŒ ๋‚˜๋ˆ ์ค„ ์ˆ˜ ์žˆ์œผ๋ฏ€๋กœ, ๋‘ ์ข…๋ฅ˜์˜ ๋ฌธ๊ตฌ๋ฅผ ํ•จ๊ป˜ ๊ณ ๋ คํ•  ๋•Œ๋Š” ์ด๋ณด๋‹ค ์ ์€ ์ˆ˜์˜ ํ•™์ƒ์—๊ฒŒ ๋‚˜๋ˆ ์ฃผ์–ด์•ผ ํ•ฉ๋‹ˆ๋‹ค. 
  - ์ด ์ค‘์—์„œ ๊ณตํ†ต์ธ์ˆ˜(common divisor)๋ฅผ ์ฐพ์•„์„œ ์ด๋ฅผ ์ตœ์†Œ ๋‹จ์œ„๋กœ ์„ค์ •ํ•˜๋ฉด ๋ฉ๋‹ˆ๋‹ค.
  - 60๊ณผ 45์˜ ์ตœ๋Œ€๊ณต์•ฝ์ˆ˜(GCD)๋Š” 15์ž…๋‹ˆ๋‹ค.
  4. ์—ฐํ•„๊ณผ ์ง€์šฐ๊ฐœ ๊ฐ๊ฐ์„ GCD(15)๋กœ ๋‚˜๋ˆ„์–ด ๋ช‡ ๊ฐœ์”ฉ ๋‚˜๋ˆ ์ค„ ์ˆ˜ ์žˆ๋Š”์ง€ ๊ตฌํ•ฉ๋‹ˆ๋‹ค.
  - ์—ฐํ•„์€ 60 / 15 = 4๊ฐœ์”ฉ, ์ด 4 * 12 = 48๊ฐœ์˜ ์—ฐํ•„์ด ํ•„์š”ํ•ฉ๋‹ˆ๋‹ค.
  - ์ง€์šฐ๊ฐœ๋Š” 45 / 15 = 3๊ฐœ์”ฉ, ์ด 3 * 1 = 3๊ฐœ์˜ ์ง€์šฐ๊ฐœ๊ฐ€ ํ•„์š”ํ•ฉ๋‹ˆ๋‹ค.
  5. ์ตœ์ข…์ ์œผ๋กœ, ๊ฐ€์žฅ ๋งŽ์€ ํ•™์ƒ๋“ค์—๊ฒŒ ๋˜‘๊ฐ™์ด ๋‚˜๋ˆ„์–ด ์ค„ ์ˆ˜ ์žˆ๋Š” ๋ฐฉ๋ฒ•์€ ์—ฐํ•„๊ณผ ์ง€์šฐ๊ฐœ๋ฅผ GCD(15)๋กœ ๋‚˜๋ˆ„์–ด ๊ฐ ํ•™์ƒ์—๊ฒŒ 4๊ฐœ์˜ ์—ฐํ•„๊ณผ 3๊ฐœ์˜ ์ง€์šฐ๊ฐœ๋ฅผ ์ œ๊ณตํ•˜๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค.
  - ์ด ๊ฒฝ์šฐ ์ด 48 + 3 = 51๊ฐœ์˜ ๋ฌผํ’ˆ์„ ์‚ฌ์šฉํ•˜์—ฌ 15๋ช…์˜ ํ•™์ƒ๋“ค์—๊ฒŒ ๋˜‘๊ฐ™์ด ๋‚˜๋ˆ ์ค„ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
    ๊ฒฐ๋ก : ์—ฐํ•„๊ณผ ์ง€์šฐ๊ฐœ ๊ฐ๊ฐ์„ GCD(15)๋กœ ๋‚˜๋ˆ„์–ด ๊ฐ ํ•™์ƒ์—๊ฒŒ 4๊ฐœ์˜ ์—ฐํ•„๊ณผ 3๊ฐœ์˜ ์ง€์šฐ๊ฐœ๋ฅผ ์ œ๊ณตํ•˜๋Š” ๋ฐฉ๋ฒ•์œผ๋กœ ์ตœ๋Œ€ 15๋ช…์˜ ํ•™์ƒ๋“ค์—๊ฒŒ ๋˜‘๊ฐ™์ด ๋‚˜๋ˆ„์–ด ์ค„ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
Writting[CLICK TO EXPAND]
User:
  ์•„๋ž˜ ๋ฌธ์žฅ์„ ๋ณด๊ณ  ๋ฌธ๋ฒ•์ ์œผ๋กœ ํ‹€๋ฆฐ๋ถ€๋ถ„์„ ์ง€์ ํ•ด์ค˜
  ๊ทธ๋Š” ์„œ์Šด์น˜ ์•Š๊ณ  ๋น„ํ‰ํ•œ๋‹ค.
Model:
  ์„œ์Šด์น˜"๋ผ๋Š” ๋‹จ์–ด๋Š” "์„œ์Šด"์ด๋ผ๋Š” ๋™์‚ฌ์™€ "์น˜"๋ผ๋Š” ์ ‘๋ฏธ์‚ฌ๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ๋งŒ๋“  ๋‹จ์–ด์ž…๋‹ˆ๋‹ค. ์ด ๋‹จ์–ด๋Š” "์„œ์Šด"์˜ ์˜๋ฏธ๋ฅผ ๋ฐ˜๋ณต์ ์œผ๋กœ ํ‘œํ˜„ํ•˜๋Š” ๋ฐ ์‚ฌ์šฉ๋˜๋Š” ์กฐ๋™์‚ฌ์  ์šฉ๋ฒ•์œผ๋กœ, "์„œ์Šด"์ด๋ž€ ๋™์‚ฌ์˜ ๋ฐ˜๋ณต์ ์ธ ํ–‰๋™์„ ๊ฐ•์กฐํ•ฉ๋‹ˆ๋‹ค. ๋”ฐ๋ผ์„œ ๋ฌธ์žฅ์—์„œ "์„œ์Šด์น˜"๋ผ๋Š” ๋‹จ์–ด๋Š” ์ค‘๋ณต์ ์œผ๋กœ ์‚ฌ์šฉ๋˜์–ด ๋ฌธ๋ฒ•์ ์œผ๋กœ ํ‹€๋ฆฐ ๋ถ€๋ถ„์ž…๋‹ˆ๋‹ค.

  ์ •๋‹ต์€ ๋‹ค์Œ๊ณผ ๊ฐ™์Šต๋‹ˆ๋‹ค:
  ๊ทธ๋Š” ์„œ์Šด์—†์ด ๋น„ํ‰ํ•œ๋‹ค.
User:
  ์œค๋™์ฃผ ์‹œ์ธ์˜ "๋ณ„ ํ—ค๋Š” ๋ฐค"์„ ๋ณด๊ณ  ๋‹น์‹œ ์‹œ๋Œ€์ƒ์„ ๊ณ ๋ คํ•ด์„œ ํ•ด์„ํ•ด๋ด. ๊ทธ๋ฆฌ๊ณ  ์ด๋ฅผ ์ž‘๊ฐ€์˜ ๋…๋ฐฑํ˜•์‹์œผ๋กœ ๋‹ค์‹œ ์ž‘์„ฑํ•ด๋ด.
Model:
  ์œค๋™์ฃผ๋Š” 1941๋…„ ์ผ๋ณธ ์ œ๊ตญ์ฃผ์˜ ์น˜ํ•˜์—์„œ ์ด ์ž‘ํ’ˆ์„ ์“ฐ๋ฉด์„œ, ๊ฐ•์••์ ์ธ ์ƒํ™ฉ ์†์—์„œ๋„ ์ž์‹ ์˜ ์ •์ฒด์„ฑ๊ณผ ๋ฌธํ™”์— ๋Œ€ํ•œ ์• ์ฐฉ์„ ํ‘œํ˜„ํ•˜๊ณ ์ž ํ–ˆ์Šต๋‹ˆ๋‹ค. ๊ทธ๋Š” ํ•˜๋Š˜์— ๊ฐ€๋“ ์ฐฌ ๋ณ„๋“ค์„ ๋ณด๋ฉฐ ๊ทธ ์†์— ํ•˜๋‚˜ ๋‘˜ ์ƒˆ๊ฒจ์ง€๋Š” ๋ณ„๋“ค์ด ์žˆ๋‹ค๊ณ  ํ‘œํ˜„ํ–ˆ์Šต๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ๋ณ„๋“ค์€ ๊ฐ๊ฐ ์ถ”์–ต, ์‚ฌ๋ž‘, ์“ธ์“ธํ•จ, ๋™๊ฒฝ, ์‹œ ๋“ฑ ๋‹ค์–‘ํ•œ ๊ฐ์ •์„ ๋‹ด๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค.
  ์œค๋™์ฃผ๋Š” ์ด ๋ณ„๋“ค ์ค‘ ํ•˜๋‚˜์— ์–ด๋จธ๋‹ˆ๋ฅผ ๋ถ€๋ฅด๋ฉฐ ๋ฉ€๋ฆฌ ๋–จ์–ด์ ธ ์žˆ๋Š” ๊ฒƒ์„ ๊ทธ๋ฆฌ์›Œํ•˜๋ฉฐ, ์ž์‹ ์˜ ์ด๋ฆ„์ž๋ฅผ ์–ธ๋• ์œ„์— ์“ฐ๊ณ  ํ™์œผ๋กœ ๋ฎ์–ด ๋ฒ„๋ฆฐ๋‹ค๊ณ  ๋งํ•ฉ๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ํ–‰๋™์€ ์ž์‹ ์˜ ์ •์ฒด์„ฑ์„ ํ™•์ธํ•˜๊ณ , ์žƒ์–ด๋ฒ„๋ฆฐ ๊ฒƒ์„ ๋˜์ฐพ์œผ๋ ค๋Š” ๋…ธ๋ ฅ์œผ๋กœ ๋ณผ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
  ์œค๋™์ฃผ๋Š” ๋˜ํ•œ ์ž์‹ ์ด ์ฃฝ๊ณ  ๋‚˜๋ฉด ๋ฌด๋ค ์œ„์— ํŒŒ๋ž€ ์ž”๋””๊ฐ€ ํ”ผ์–ด๋‚˜๋“ฏ์ด ์ž์‹ ์˜ ์ด๋ฆ„์ž ๋ฌปํžŒ ์–ธ๋• ์œ„์—๋„ ํ’€์ด ๋ฌด์„ฑํ•  ๊ฒƒ์ด๋ผ๊ณ  ๋งํ•ฉ๋‹ˆ๋‹ค. ์ด๋Š” ๋‹น์‹œ์˜ ์‚ฌํšŒ์ƒ๊ณผ ์ž‘๊ฐ€์˜ ๋‚ด๋ฉด์„ ๋ฐ˜์˜ํ•˜๊ณ  ์žˆ์œผ๋ฉฐ, ํ•œ๊ตญ์ธ์˜ ์ •์ฒด์„ฑ์„ ํ™•์ธํ•˜๊ณ  ์žƒ์–ด๋ฒ„๋ฆฐ ๊ฒƒ์„ ๋˜์ฐพ์œผ๋ ค๋Š” ๋…ธ๋ ฅ, ๊ทธ๋ฆฌ๊ณ  ์ฃฝ์Œ ์ดํ›„์— ์ž์‹ ์˜ ์กด์žฌ๋ฅผ ๋‚จ๊ฒจ๋‘๊ณ  ์‹ถ๋‹ค๋Š” ๋ฐ”๋žŒ์„ ๋‹ด๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค.
  ์ด ์‹œ๋ฅผ ์ž‘๊ฐ€์˜ ๋…๋ฐฑํ˜•์‹์œผ๋กœ ๋‹ค์‹œ ์ž‘์„ฑํ•˜๋ฉด ๋‹ค์Œ๊ณผ ๊ฐ™์Šต๋‹ˆ๋‹ค:
  "์ €๋Š” ์ด ๋ณ„๋“ค์ด ๋„ˆ๋ฌด๋‚˜ ๋งŽ์€ ๊ฒƒ ๊ฐ™์•„์š”. ํ•˜๋‚˜ ๋‘˜ ์ƒˆ๊ฒจ์ง€๋Š” ๋ณ„๋“ค ์ค‘์—๋Š” ์ถ”์–ต๋„ ์žˆ๊ณ , ์‚ฌ๋ž‘๋„ ์žˆ๊ณ , ์“ธ์“ธํ•จ๊ณผ ๋™๊ฒฝ๋„ ์žˆ์–ด์š”. ๊ทธ๋ฆฌ๊ณ  ์–ด๋จธ๋‹ˆ, ๋‹น์‹ ์€ ๋ฉ€๋ฆฌ ๋ถ๊ฐ„๋„์— ๊ณ„์‹  ๊ฑด๊ฐ€์š”? ์ €๋Š” ๋‹น์‹ ์„ ๋ถ€๋ฅด๊ณ  ์‹ถ์–ด์š”.
  ์ €์˜ ์ด๋ฆ„์ž๋Š” ์–ธ๋• ์œ„์— ์“ฐ๊ณ  ํ™์œผ๋กœ ๋ฎ์–ด ๋ฒ„๋ ธ์–ด์š”. ๋”ด์€ ๋ฐค์„ ์ƒˆ์›Œ ์šฐ๋Š” ๋ฒŒ๋ ˆ์ฒ˜๋Ÿผ ๋ถ€๋„๋Ÿฌ์šด ์ด๋ฆ„์„ ์Šฌํผํ•˜๋Š” ์ด์œ ๊ฐ€ ์žˆ์–ด์š”. ํ•˜์ง€๋งŒ ๊ฒจ์šธ์ด ์ง€๋‚˜๊ณ  ๋‚˜์˜ ๋ณ„์—๋„ ๋ด„์ด ์˜ค๋ฉด, ๋ฌด๋ค ์œ„์— ํŒŒ๋ž€ ์ž”๋””๊ฐ€ ํ”ผ์–ด๋‚˜๋“ฏ์ด ์ €์˜ ์ด๋ฆ„์ž ๋ฌปํžŒ ์–ธ๋• ์œ„์—์„œ๋„ ํ’€์ด ๋ฌด์„ฑํ•  ๊ฑฐ์˜ˆ์š”.
  ์ €๋Š” ์ด ๋ณ„๋“ค์„ ๋ณด๊ณ , ์ž์‹ ์˜ ์ •์ฒด์„ฑ์„ ํ™•์ธํ•˜๊ณ  ์‹ถ์–ด์š”. ์žƒ์–ด๋ฒ„๋ฆฐ ๊ฒƒ์„ ๋˜์ฐพ์œผ๋ ค๋Š” ๋…ธ๋ ฅ์„ ํ•˜๊ณ  ์‹ถ์–ด์š”. ์ฃฝ์Œ ์ดํ›„์—๋„ ์ž์‹ ์˜ ์กด์žฌ๋ฅผ ๋‚จ๊ฒจ๋‘๊ณ  ์‹ถ์–ด์š”."
Logical Reasoning[CLICK TO EXPAND]
User:
  ์„ธ์ข…๋Œ€์™•์˜ ๋งฅ๋ถํ”„๋กœ ๋˜์ง ์‚ฌ๊ฑด์— ๋Œ€ํ•ด ์„ค๋ช…ํ•ด์ค˜
Model:
  ์„ธ์ข…๋Œ€์™•์ด ๋งฅ๋ถํ”„๋กœ๋ฅผ ๋˜์ง„ ์‚ฌ๊ฑด์€ ์‚ฌ์‹ค๋ฌด๊ทผํ•œ ์ด์•ผ๊ธฐ์ž…๋‹ˆ๋‹ค. ์„ธ์ข…๋Œ€์™•์€ ์กฐ์„  ์ œ4๋Œ€ ๊ตญ์™•์œผ๋กœ 1418๋…„๋ถ€ํ„ฐ 1450๋…„๊นŒ์ง€ ์žฌ์œ„ํ–ˆ์Šต๋‹ˆ๋‹ค. ๊ทธ๋Š” ๊ณผํ•™, ๋ฌธํ•™, ์˜ˆ์ˆ  ๋“ฑ ๋‹ค์–‘ํ•œ ๋ถ„์•ผ์— ๊ด€์‹ฌ์„ ๊ฐ–๊ณ  ์—ฐ๊ตฌํ•˜์˜€์œผ๋ฉฐ, ํŠนํžˆ ํ•œ๊ธ€ ์ฐฝ์ œ์™€ ์ฒœ๋ฌธํ•™์— ํฐ ๊ณตํ—Œ์„ ํ–ˆ์Šต๋‹ˆ๋‹ค.
  ํ•œํŽธ ๋งฅ๋ถํ”„๋กœ๋Š” ์• ํ”Œ์ด ์ƒ์‚ฐํ•˜๋Š” ๋…ธํŠธ๋ถ ์ปดํ“จํ„ฐ๋กœ, ์„ธ์ข…๋Œ€์™• ์‹œ๋Œ€์˜ ๊ธฐ์ˆ ๋กœ๋Š” ์กด์žฌํ•  ์ˆ˜ ์—†๋Š” ์ œํ’ˆ์ž…๋‹ˆ๋‹ค. ๋”ฐ๋ผ์„œ ์„ธ์ข…๋Œ€์™•์ด ๋งฅ๋ถํ”„๋กœ๋ฅผ ๋˜์ง„ ์‚ฌ๊ฑด์€ ์‚ฌ์‹ค๊ณผ ๋งž์ง€ ์•Š๋Š” ์ด์•ผ๊ธฐ๊ฐ€ ๋  ๊ฒƒ์ž…๋‹ˆ๋‹ค.
Coding [CLICK TO EXPAND]
User:
  PyTorch๋ฅผ ์‚ฌ์šฉํ•ด์„œ ๊ฐ„๋‹จํ•œ CNN์„ ๊ตฌํ˜„ํ•ด๋ด.
Model:
  ๋„ค, ๋ฌผ๋ก ์ด์—์š”! PyTorch๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ๊ฐ„๋‹จํ•œ CNN์„ ๊ตฌํ˜„ํ•˜๋Š” ์˜ˆ์ œ ์ฝ”๋“œ๋ฅผ ๋ณด์—ฌ๋“œ๋ฆด๊ฒŒ์š”.
  ์šฐ์„ , ํ•„์š”ํ•œ ๋ผ์ด๋ธŒ๋Ÿฌ๋ฆฌ๋ฅผ ๊ฐ€์ ธ์™€์•ผ ํ•ด์š”. PyTorch์˜ `torch`์™€ `nn` ๋ชจ๋“ˆ์ด ํ•„์š”ํ•ด์š”. 
```python
import torch\nimport torch.nn as nn
```
๊ทธ ๋‹ค์Œ, ๊ฐ„๋‹จํ•œ CNN ๋ชจ๋ธ์„ ์ •์˜ํ•  ๊ฑฐ์˜ˆ์š”. ์ด ๊ฒฝ์šฐ์—๋Š” 2๊ฐœ์˜ ์ปจ๋ณผ๋ฃจ์…˜ ๋ ˆ์ด์–ด์™€ 3๊ฐœ์˜ ํ’€๋ง ๋ ˆ์ด์–ด๋กœ ๊ตฌ์„ฑ๋œ ๋ชจ๋ธ์ด์—์š”.
  ```python
class SimpleCNN(nn.Module):
def __init__(self):
  super(SimpleCNN, self).__init__()
  self.conv1 = nn.Conv2d(1, 10, kernel_size=5)
  self.pool = nn.MaxPool2d(2, 2)
  self.conv2 = nn.Conv2d(10, 20, kernel_size=5)

def forward(self, x):
  x = self.pool(torch.relu(self.conv1(x)))
  x = self.pool(torch.relu(self.conv2(x)))
  return x
```
์ด์ œ ๋ชจ๋ธ์„ ์ดˆ๊ธฐํ™”ํ•˜๊ณ , ์ž…๋ ฅ ๋ฐ์ดํ„ฐ๋ฅผ ์ค€๋น„ํ•ด์•ผ ํ•ด์š”. ์—ฌ๊ธฐ์„œ๋Š” ์ž„์˜์˜ 28x28 ํฌ๊ธฐ์˜ ์ด๋ฏธ์ง€ ๋ฐ์ดํ„ฐ๋ฅผ ์ƒ์„ฑํ–ˆ์–ด์š”.
```python
model = SimpleCNN()
input_data = torch.randn(1, 1, 28, 28)
```
๋งˆ์ง€๋ง‰์œผ๋กœ, ๋ชจ๋ธ์— ์ž…๋ ฅ ๋ฐ์ดํ„ฐ๋ฅผ ๋„ฃ์–ด ์ถœ๋ ฅ์„ ํ™•์ธํ•  ์ˆ˜ ์žˆ์–ด์š”.
```python
output = model(input_data)
print(output.shape)
```
์ด๋ ‡๊ฒŒ ํ•˜๋ฉด ๊ฐ„๋‹จํ•œ CNN ๋ชจ๋ธ์„ PyTorch๋ฅผ ์‚ฌ์šฉํ•ด ๊ตฌํ˜„ํ•˜๊ณ , ์ž…๋ ฅ ๋ฐ์ดํ„ฐ๋ฅผ ํ†ตํ•ด ์ถœ๋ ฅ์„ ํ™•์ธํ•  ์ˆ˜ ์žˆ์–ด์š”. ๋ฌผ๋ก  ์‹ค์ œ๋กœ๋Š” ๋” ๋งŽ์€ ๋ ˆ์ด์–ด๋‚˜ ํŒŒ๋ผ๋ฏธํ„ฐ ํŠœ๋‹ ๋“ฑ์ด ํ•„์š”ํ•˜์ง€๋งŒ, ์ด ์ฝ”๋“œ๋Š” ๊ธฐ๋ณธ์ ์ธ CNN ๊ตฌ์กฐ๋ฅผ ๋ณด์—ฌ์ฃผ๋Š” ์˜ˆ์ œ์ผ ๋ฟ์ด์—์š”. ๋” ๊ถ๊ธˆํ•œ ์ ์ด ์žˆ์œผ๋ฉด ์–ธ์ œ๋“ ์ง€ ๋ฌผ์–ด๋ด ์ฃผ์„ธ์š”!

You can find more examples at our project page

Warning

Although AKALlama-70B has significant potential, its responses can sometimes be inaccurate, biased, or misaligned, presenting risks if used without additional testing and refinement. Furthermore, the quality of the model's output is greatly influenced by the system prompt and decoding strategy. Changes in these areas could result in less precise outputs. Therefore, we strongly recommend handling our model with considerable caution.

Citation

If you find our model useful in your work, please cite the model as follows:

@misc{akallama,
  author = {Chung, Jiwan and Jeon, Jaehyun and Kim, Saejin and Lim, Seungwon and Oh, Giyeong and Son, Yejin and Yu, Youngjae},
  title = {AkaLlama: Yonsei University Large Language Model Project},
  year = {2024},
  publisher = {Hugging Face},
  journal = {Hugging Face repository},
  howpublished = {\url{https://huggingface.co/mirlab/AkaLlama-llama3-70b-v0.1}},
}

Contact

We look forward for your feedback and welcome collaboration on this exciting project!

Contributors

Special Thanks

  • Data Center of the Department of Artificial Intelligence at Yonsei University for the computation resources

Acknowledgement

  • Title image generated by DALLยทE 3
Downloads last month
985
GGUF
Model size
25.1B params
Architecture
undefined

2-bit

3-bit

4-bit

5-bit

6-bit

8-bit

Inference Examples
Inference API (serverless) has been turned off for this model.

Collection including mirlab/AkaLlama-llama3-70b-v0.1-GGUF