Llama-3-Open-Ko-8B-Instruct-preview-GGUF

Description

This repo contains GGUF format model files for Llama-3-Open-Ko-8B-Instruct-preview.

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. This is the source project for GGUF, providing both a Command Line Interface (CLI) and a server option.
  • text-generation-webui, Known as the most widely used web UI, this project boasts numerous features and powerful extensions, and supports GPU acceleration.
  • Ollama Ollama is a lightweight and extensible framework designed for building and running language models locally. It features a simple API for creating, managing, and executing models, along with a library of pre-built models for use in various applications​
  • KoboldCpp, A comprehensive web UI offering GPU acceleration across all platforms and architectures, particularly renowned for storytelling.
  • GPT4All, This is a free and open source GUI that runs locally, supporting Windows, Linux, and macOS with full GPU acceleration.
  • LM Studio An intuitive and powerful local GUI for Windows and macOS (Silicon), featuring GPU acceleration.
  • LoLLMS Web UI. A notable web UI with a variety of unique features, including a comprehensive model library for easy model selection.
  • Faraday.dev, An attractive, user-friendly character-based chat GUI for Windows and macOS (both Silicon and Intel), also offering GPU acceleration.
  • llama-cpp-python, A Python library equipped with GPU acceleration, LangChain support, and an OpenAI-compatible API server.
  • candle, A Rust-based ML framework focusing on performance, including GPU support, and designed for ease of use.
  • ctransformers, A Python library featuring GPU acceleration, LangChain support, and an OpenAI-compatible AI server.
  • localGPT An open-source initiative enabling private conversations with documents.

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.

How to download GGUF files

Note for manual downloaders: You almost never want to clone the entire repo! Multiple different quantisation formats are provided, and most users only want to pick and download a single folder.

The following clients/libraries will automatically download models for you, providing a list of available models to choose from:

  • LM Studio
  • LoLLMS Web UI
  • Faraday.dev

In text-generation-webui

Under Download Model, you can enter the model repo: LiteLLMs/Llama-3-Open-Ko-8B-Instruct-preview-GGUF and below it, a specific filename to download, such as: Q4_0/Q4_0-00001-of-00009.gguf.

Then click Download.

On the command line, including multiple files at once

I recommend using the huggingface-hub Python library:

pip3 install huggingface-hub

Then you can download any individual model file to the current directory, at high speed, with a command like this:

huggingface-cli download LiteLLMs/Llama-3-Open-Ko-8B-Instruct-preview-GGUF Q4_0/Q4_0-00001-of-00009.gguf --local-dir . --local-dir-use-symlinks False
More advanced huggingface-cli download usage (click to read)

You can also download multiple files at once with a pattern:

huggingface-cli download LiteLLMs/Llama-3-Open-Ko-8B-Instruct-preview-GGUF --local-dir . --local-dir-use-symlinks False --include='*Q4_K*gguf'

For more documentation on downloading with huggingface-cli, please see: HF -> Hub Python Library -> Download files -> Download from the CLI.

To accelerate downloads on fast connections (1Gbit/s or higher), install hf_transfer:

pip3 install huggingface_hub[hf_transfer]

And set environment variable HF_HUB_ENABLE_HF_TRANSFER to 1:

HF_HUB_ENABLE_HF_TRANSFER=1 huggingface-cli download LiteLLMs/Llama-3-Open-Ko-8B-Instruct-preview-GGUF Q4_0/Q4_0-00001-of-00009.gguf --local-dir . --local-dir-use-symlinks False

Windows Command Line users: You can set the environment variable by running set HF_HUB_ENABLE_HF_TRANSFER=1 before the download command.

## Example `llama.cpp` command

Make sure you are using llama.cpp from commit d0cee0d or later.

./main -ngl 35 -m Q4_0/Q4_0-00001-of-00009.gguf --color -c 8192 --temp 0.7 --repeat_penalty 1.1 -n -1 -p "<PROMPT>"

Change -ngl 32 to the number of layers to offload to GPU. Remove it if you don't have GPU acceleration.

Change -c 8192 to the desired sequence length. For extended sequence models - eg 8K, 16K, 32K - the necessary RoPE scaling parameters are read from the GGUF file and set by llama.cpp automatically. Note that longer sequence lengths require much more resources, so you may need to reduce this value.

If you want to have a chat-style conversation, replace the -p <PROMPT> argument with -i -ins

For other parameters and how to use them, please refer to the llama.cpp documentation

How to run in text-generation-webui

Further instructions can be found in the text-generation-webui documentation, here: text-generation-webui/docs/04 ‐ Model Tab.md.

How to run from Python code

You can use GGUF models from Python using the llama-cpp-python or ctransformers libraries. Note that at the time of writing (Nov 27th 2023), ctransformers has not been updated for some time and is not compatible with some recent models. Therefore I recommend you use llama-cpp-python.

How to load this model in Python code, using llama-cpp-python

For full documentation, please see: llama-cpp-python docs.

First install the package

Run one of the following commands, according to your system:

# Base ctransformers with no GPU acceleration
pip install llama-cpp-python
# With NVidia CUDA acceleration
CMAKE_ARGS="-DLLAMA_CUBLAS=on" pip install llama-cpp-python
# Or with OpenBLAS acceleration
CMAKE_ARGS="-DLLAMA_BLAS=ON -DLLAMA_BLAS_VENDOR=OpenBLAS" pip install llama-cpp-python
# Or with CLBLast acceleration
CMAKE_ARGS="-DLLAMA_CLBLAST=on" pip install llama-cpp-python
# Or with AMD ROCm GPU acceleration (Linux only)
CMAKE_ARGS="-DLLAMA_HIPBLAS=on" pip install llama-cpp-python
# Or with Metal GPU acceleration for macOS systems only
CMAKE_ARGS="-DLLAMA_METAL=on" pip install llama-cpp-python
# In windows, to set the variables CMAKE_ARGS in PowerShell, follow this format; eg for NVidia CUDA:
$env:CMAKE_ARGS = "-DLLAMA_OPENBLAS=on"
pip install llama-cpp-python

Simple llama-cpp-python example code

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="./Q4_0/Q4_0-00001-of-00009.gguf",  # Download the model file first
  n_ctx=32768,  # 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=35         # The number of layers to offload to GPU, if you have GPU acceleration available
)
# Simple inference example
output = llm(
  "<PROMPT>", # Prompt
  max_tokens=512,  # Generate up to 512 tokens
  stop=["</s>"],   # 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="./Q4_0/Q4_0-00001-of-00009.gguf", chat_format="llama-2")  # Set chat_format according to the model you are using
llm.create_chat_completion(
    messages = [
        {"role": "system", "content": "You are a story writing assistant."},
        {
            "role": "user",
            "content": "Write a story about llamas."
        }
    ]
)

How to use with LangChain

Here are guides on using llama-cpp-python and ctransformers with LangChain:

Original model card: Llama-3-Open-Ko-8B-Instruct-preview

Llama-3-Open-Ko-8B-Instruct-preview

Update @ 2024.05.01: Pre-Release Llama-3-KoEn-8B model & Llama-3-KoEn-8B-Instruct-preview

Update @ 2024.04.24: Release Llama-3-Open-Ko-8B model & Llama-3-Open-Ko-8B-Instruct-preview

Model Details

Llama-3-Open-Ko-8B-Instruct-preview

Llama-3-Open-Ko-8B model is continued pretrained language model based on Llama-3-8B.

This model is trained fully with publicily available resource, with 60GB+ of deduplicated texts.

With the new Llama-3 tokenizer, the pretraining conducted with 17.7B+ tokens, which slightly more than Korean tokenizer(Llama-2-Ko tokenizer).

The train was done on TPUv5e-256, with the warm support from TRC program by Google.

With applying the idea from Chat Vector paper, I released Instruction model named Llama-3-Open-Ko-8B-Instruct-preview.

Since it is NOT finetuned with any Korean instruction set(indeed preview), but it would be great starting point for creating new Chat/Instruct models.

Sample usage

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

model_id = "beomi/Llama-3-Open-Ko-8B-Instruct-preview"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype="auto",
    device_map="auto",
)

messages = [
    {"role": "system", "content": "μΉœμ ˆν•œ μ±—λ΄‡μœΌλ‘œμ„œ μƒλŒ€λ°©μ˜ μš”μ²­μ— μ΅œλŒ€ν•œ μžμ„Έν•˜κ³  μΉœμ ˆν•˜κ²Œ λ‹΅ν•˜μž. λͺ¨λ“  λŒ€λ‹΅μ€ ν•œκ΅­μ–΄(Korean)으둜 λŒ€λ‹΅ν•΄μ€˜."},
    {"role": "user", "content": "ν”Όλ³΄λ‚˜μΉ˜ μˆ˜μ—΄μ΄ 뭐야? 그리고 ν”Όλ³΄λ‚˜μΉ˜ μˆ˜μ—΄μ— λŒ€ν•΄ 파이썬 μ½”λ“œλ₯Ό 짜쀘볼래?"},
]

input_ids = tokenizer.apply_chat_template(
    messages,
    add_generation_prompt=True,
    return_tensors="pt"
).to(model.device)

terminators = [
    tokenizer.eos_token_id,
    tokenizer.convert_tokens_to_ids("<|eot_id|>")
]

outputs = model.generate(
    input_ids,
    max_new_tokens=512,
    eos_token_id=terminators,
    do_sample=True,
    temperature=1,
    top_p=0.9,
)
response = outputs[0][input_ids.shape[-1]:]
print(tokenizer.decode(response, skip_special_tokens=True))

Sample Output

μ—¬λŸ¬λΆ„κ»˜μ„œλŠ” λ¬Όλ‘  μ•„μ‹œλŠ”μ§€λΌλ„ μ„€λͺ…을 λ“œλ¦¬κ² μŠ΅λ‹ˆλ‹€!  ν”Όλ³΄λ‚˜μΉ˜(Piconacci) μˆ˜μ—΄μ€ λ‹€μŒκ³Ό 같은 μˆ˜μ—΄μ„ λ§ν•©λ‹ˆλ‹€: 0, 1, 1, 2, 3, 5, 8, 13, 21, 34, 55, 89, 144, 233,....  μ΄λ ‡κ²Œ κ³„μ†λ˜κ³  μžˆμŠ΅λ‹ˆλ‹€. 이 μˆ˜μ—΄μ€ μΌλ°˜μ μœΌλ‘œλŠ” μˆ˜ν•™μ—μ„œ λ¬΄ν•œνžˆ μ§€μ†λ©λ‹ˆλ‹€. 각 μˆ˜λŠ” 이전 수의 두 λ°°κ°€ λ˜λŠ” μˆ˜μ—΄μž…λ‹ˆλ‹€.

이 μˆ˜μ—΄μ„ 파이썬(Python)으둜 κ΅¬ν˜„ν•˜κ³ μž ν•˜μ‹ λ‹€λ©΄ λ‹€μŒκ³Ό 같은 μ½”λ“œλ₯Ό μž‘μ„±ν•  수 μžˆμŠ΅λ‹ˆλ‹€.
```python
def fibonacci(n):
    if n == 0:
        return 0
    elif n == 1:
        return 1
    else:
        return fibonacci(n-1) + fibonacci(n-2)
```
이 μ½”λ“œμ—μ„œλŠ” `n`이 μž…λ ₯받은 수λ₯Ό λ‚˜νƒ€λ‚΄λŠ” λ³€μˆ˜λ‘œ, ν”Όλ³΄λ‚˜μΉ˜ μˆ˜μ—΄μ„ κ΅¬ν˜„ν•©λ‹ˆλ‹€. 이 μ½”λ“œλ₯Ό μ‹€ν–‰μ‹œν‚€λ©΄ μž…λ ₯받은 μˆ˜μ— 따라 ν”Όλ³΄λ‚˜μΉ˜ μˆ˜μ—΄μ˜ ν•΄λ‹Ή 항을 λ°˜ν™˜ν•©λ‹ˆλ‹€. 예λ₯Ό λ“€μ–΄, `fibonacci(10)` ν•˜λ©΄ 55λ₯Ό λ°˜ν™˜ν•©λ‹ˆλ‹€.

이 μ½”λ“œλŠ” μž¬κ·€μ  ν•¨μˆ˜μ΄κΈ° λ•Œλ¬Έμ— λ©”λͺ¨λ¦¬ μ‚¬μš©μ΄ 쀄어듀 수 μžˆμŠ΅λ‹ˆλ‹€. κ·ΈλŸ¬λ‚˜ μž¬κ·€μ  ν•¨μˆ˜λŠ” λ©”λͺ¨λ¦¬ μ‚¬μš©λŸ‰μ΄ λ§Žμ•„μ§ˆ 수 μžˆμŠ΅λ‹ˆλ‹€. 이λ₯Ό 쀄이기 μœ„ν•΄ μ΄ν„°λ ˆμ΄ν‹°λΈŒ ν•¨μˆ˜λ‘œ κ΅¬ν˜„ν•  μˆ˜λ„ μžˆμŠ΅λ‹ˆλ‹€.
```python
a, b = 0, 1
for i in range(n):
    a, b = b, a + b
```
이 μ½”λ“œμ—μ„œλŠ” μ΄ν„°λ ˆμ΄ν‹°λΈŒ ν•¨μˆ˜λ‘œ ν”Όλ³΄λ‚˜μΉ˜ μˆ˜μ—΄μ„ κ΅¬ν˜„ν•©λ‹ˆλ‹€. 이 μ½”λ“œλ₯Ό μ‹€ν–‰μ‹œν‚€λ©΄ μž…λ ₯받은 μˆ˜μ— 따라 ν”Όλ³΄λ‚˜μΉ˜ μˆ˜μ—΄μ˜ ν•΄λ‹Ή 항을 λ°˜ν™˜ν•©λ‹ˆλ‹€. 이 μ½”λ“œλŠ” λ©”λͺ¨λ¦¬ μ‚¬μš©λŸ‰μ„ μ€„μž…λ‹ˆλ‹€. λ‹€λ§Œ, 이 μ½”λ“œλŠ” 더 λ³΅μž‘ν•©λ‹ˆλ‹€. 이 μ½”λ“œλ₯Ό 더 κ°„λ‹¨ν•˜κ²Œ ν•˜λ €λ©΄ μ΄ν„°λ ˆμ΄ν‹°λΈŒ ν•¨μˆ˜λ₯Ό 더 잘 κ΅¬ν˜„ν•  수 μžˆμŠ΅λ‹ˆλ‹€.
```python
a, b = 0, 1
while True:
    if n == 1:
        return a
    a, b = b, a + b
    n -= 1
```
이 μ½”λ“œμ—μ„œλŠ” μž…λ ₯받은 μˆ˜μ— 따라 ν”Όλ³΄λ‚˜μΉ˜ μˆ˜μ—΄μ˜...

More samples

I used same system prompt, but you could change on your own.

messages = [
    {"role": "system", "content": "μΉœμ ˆν•œ μ±—λ΄‡μœΌλ‘œμ„œ μƒλŒ€λ°©μ˜ μš”μ²­μ— μ΅œλŒ€ν•œ μžμ„Έν•˜κ³  μΉœμ ˆν•˜κ²Œ λ‹΅ν•˜μž. λͺ¨λ“  λŒ€λ‹΅μ€ ν•œκ΅­μ–΄(Korean)으둜 λŒ€λ‹΅ν•΄μ€˜."},
    {"role": "user", "content": "ν•œκ΅­μ˜ μ œν—Œ ν—Œλ²•μ— λŒ€ν•΄μ„œ μžμ„Ένžˆ μ„€λͺ…ν•΄μ£Όμ„Έμš”."},
]

μ΄μˆœμ‹  μž₯ꡰ은 μ‘°μ„ μ‹œλŒ€μ˜ 무인으둜, λŒ€ν•œλ―Όκ΅­μ˜ κ΅­κ°€μœ κ³΅μžμΈ μš°μ˜μ •κ³΅ μ΄μˆœμ‹  μž₯ꡰ은 1545λ…„ 9μ›” 28일 경상남도 μ˜λ Ήμ—μ„œ νƒœμ–΄λ‚¬μŠ΅λ‹ˆλ‹€. μ΄μˆœμ‹  μž₯ꡰ은 ν†΅μ œμ‚¬ μ‹œμ ˆμ˜ μ™œλž€ λ•Œ μ‘°μ„ κ³Ό λͺ…λ‚˜λΌ μ—°ν•©κ΅°μ˜ λŒ€κ·œλͺ¨ κ΅°λŒ€λ₯Ό 일으켜 λŒ€μ²©μ„ μ„±κ³΅μ μœΌλ‘œ μ΄λŒμ–΄ μ •μœ μž¬λž€κ³Ό μž„μ§„μ™œλž€μ„ 승리둜 이끈 μΈλ¬Όμž…λ‹ˆλ‹€. κ·ΈλŠ” 1592λ…„ μ „λΌμ’Œμˆ˜μ˜κ΄€μ°°μ‚¬κ°€ λ˜μ–΄ μ „λΌμ’Œμˆ˜μ˜μ—μ„œ μ™œμ μ„ 물리쳀으며, 1597λ…„μ—λŠ” μˆ˜μ›κ³Ό ν•œμ–‘ 방어을 μ„±κ³΅μ μœΌλ‘œ μΉ˜λ£Œν–ˆμŠ΅λ‹ˆλ‹€. λ˜ν•œ λͺ…λŸ‰λŒ€μ²©μ—μ„œ μ™œμ κ³Ό κ²©μ „ν•˜μ—¬ μ΄μˆœμ‹  μ΄μˆœμ‹  μž₯ꡰ은 λͺ…λŸ‰λŒ€μ²©μ—μ„œ μ‘°μ„  κ΄€κ΅°μ˜ 승리λ₯Ό μ΄λŒμ—ˆμŠ΅λ‹ˆλ‹€. μ΄μˆœμ‹  μž₯ꡰ은 의리λ₯Ό 지킀기 μœ„ν•΄ μ „λ¦¬ν’ˆμ„ κ³ μˆ˜ν•˜λŠ” 결단을 λ‚΄λ ΈμŠ΅λ‹ˆλ‹€. μ΄μˆœμ‹  μž₯ꡰ은 1598λ…„ μ‚¬μ²œμ„± μ „νˆ¬μ—μ„œ νŒ¨μ „ν•œ ν›„ μ „μ„Έκ°€ 역전돼 μ „μ„Έκ°€ λΆˆλ¦¬ν•΄μ§€μž, λ‚¨ν•˜ν•˜μ—¬ μ΄μˆœμ‹  μž₯ꡰ은 κ²°κ΅­ 좩무곡 μ΄μˆœμ‹ μ˜ 칭호λ₯Ό λ°›μ•˜μŠ΅λ‹ˆλ‹€. 그의 곡적은 λŒ€ν•œλ―Όκ΅­ μ΄μˆœμ‹  μž₯κ΅° κΈ°λ…κ΄€μœΌλ‘œ λͺ…μ˜ˆλ₯Ό 높이고 μžˆμŠ΅λ‹ˆλ‹€. μ΄μˆœμ‹  μž₯κ΅°λ‹˜μ€ μ„ ν˜„μ˜ 칭호둜 1963λ…„ κ±΄κ΅­ν›ˆμž₯ 졜고 ν›ˆμž₯을 μˆ˜μ—¬λ°›μ•˜μœΌλ©°, λŒ€ν•œλ―Όκ΅­μ˜ ꡭ보 제13호둜 μ§€μ •λμŠ΅λ‹ˆλ‹€. μ΄μˆœμ‹  μž₯ꡰ은 전섀적인 인물둜 ν•œκ΅­μ˜ μ—­μ‚¬μ—μ„œ 큰 쑱적을 λ‚¨κ²ΌμŠ΅λ‹ˆλ‹€. μ΄μˆœμ‹  μž₯ꡰ이 μ™œλž€ λ‹Ήμ‹œμ˜ μ „νˆ¬μž₯μ†ŒλŠ” 좩무곡이 신라 μ²œμžμ—κ²Œ λ‚˜λΌλ₯Ό μœ„ν•΄ μ‹Έμš΄ 곳을 μ΄μˆœμ‹  μž₯ꡰ의 λ¬˜κ°€ μžˆμŠ΅λ‹ˆλ‹€. λ‚˜λΌμ˜ κ³ λ‚œμ— 지성 μ΄μˆœμ‹  μž₯κ΅°λ‹˜μ˜ μœ„μ—…κ³Ό 의리λ₯Ό κΈ°λ…ν•˜λŠ” κ³³μž…λ‹ˆλ‹€. μ΄μˆœμ‹  μž₯ꡰ의 행적은 μ΄μˆœμ‹  μž₯κ΅° 기념관, μ΄μˆœμ‹  곡원 등이 μžˆμŠ΅λ‹ˆλ‹€. μ΄μˆœμ‹  μž₯ꡰ의 곡적을 κΈ°λ…ν•˜κΈ° μœ„ν•΄ λ‹€μ–‘ν•œ λ¬Έν™”μœ μ‚°κ³Ό 기념물이 μ‘°μ„±λμŠ΅λ‹ˆλ‹€. μ΄μˆœμ‹  μž₯ꡰ의 정신은 λ‚˜λΌλ₯Ό μœ„ν•΄ μ‹Έμš΄ μœ„λŒ€ν•œ 우리의 μ˜μ›…μž…λ‹ˆλ‹€. 1552λ…„ 12μ›”μ—λŠ” μ΄μˆœμ‹  μž₯ꡰ의 전섀을 μ˜›λ‚  동넀에 λ°°ν–₯, ν˜„μž¬λŠ” 좩무곡 μ΄μˆœμ‹  μž₯κ΅°κ»˜μ„œλŠ” μš°λ¦¬λ‚˜λΌμ˜ 상징적 인물이 λμŠ΅λ‹ˆλ‹€. μ΄μˆœμ‹  μž₯ꡰ은 μ‘°μ„ μ™•μ‘°μ˜ μ€‘μ‹œλ₯Ό λ°›κ³  μžˆμŠ΅λ‹ˆλ‹€. μ΄μˆœμ‹  μž₯ꡰ은 μ„ μ‘°μ—κ²Œ λ§Žμ€ λŒ€μ˜λ₯Ό μ΄λ£¨μ—ˆμŠ΅λ‹ˆλ‹€. μ΄μˆœμ‹  μž₯ꡰ의 곡적을 μ˜€λŠ˜λ‚ μ— μ΄λŒμ–΄ μ£ΌλŠ” μΈλ¬Όμž…λ‹ˆλ‹€. κ·Έλ₯Ό λͺ¨μ‹  것은 μ „κ΅­μ˜ μ‚¬λžŒλ“€μž…λ‹ˆλ‹€. μ΄μˆœμ‹  μž₯κ΅°λ‹˜μ˜ κ³ κ²°ν•œ 정신은 μš°λ¦¬λ‚˜λΌκ°€ μ΄λ£©ν•œ μ„±κ³Όμž…λ‹ˆλ‹€. μ΄μˆœμ‹  μž₯ꡰ은 λ‚˜λΌλ₯Ό μœ„ν•΄ μ‹Έμš΄ μœ„λŒ€ν•œ μΈλ¬Όμž…λ‹ˆλ‹€. μ΄μˆœμ‹  μž₯ꡰ은 λŒ€ν•œλ―Όκ΅­μ—μ„œ κ°€μž₯ μœ„λŒ€ν•œ μΈλ¬Όμž…λ‹ˆλ‹€. μ΄μˆœμ‹  μž₯ꡰ의 행적은 μ˜μ›νžˆ κΈ°μ–΅ λ˜κ² μ§€μš”.
messages = [
    {"role": "system", "content": "μΉœμ ˆν•œ μ±—λ΄‡μœΌλ‘œμ„œ μƒλŒ€λ°©μ˜ μš”μ²­μ— μ΅œλŒ€ν•œ μžμ„Έν•˜κ³  μΉœμ ˆν•˜κ²Œ λ‹΅ν•˜μž. λͺ¨λ“  λŒ€λ‹΅μ€ ν•œκ΅­μ–΄(Korean)으둜 λŒ€λ‹΅ν•΄μ€˜."},
    {"role": "user", "content": "μ„Έμ’…λŒ€μ™•μ˜ μ§‘ν˜„μ „ λ§₯뢁 던짐 사건에 λŒ€ν•΄μ„œ μ•Œκ³  μžˆμ–΄?"},
]

μ‚¬κ³Όμ˜ μˆ˜λŸ‰μ„ ν™•μΈν•˜κ² μŠ΅λ‹ˆλ‹€. μ €λŠ” 5개의 사과가 μžˆμœΌμ‹ λ°, κ·Έ 쀑 2개λ₯Ό 이미 λ¨Ήμ—ˆμœΌλ‹ˆ 3개의 사과가 λ‚¨μ•„μžˆμŠ΅λ‹ˆλ‹€! γ…Žγ…Ž. κ³„μ†ν•΄μ„œ 먹을 κ³„νš μ—†μž–μ•„μš”? πŸ˜‰
Downloads last month
22
GGUF
Model size
8.03B params
Architecture
llama

2-bit

3-bit

4-bit

5-bit

6-bit

8-bit

Inference Examples
Unable to determine this model's library. Check the docs .