README.md

metadata

base_model:
  - Bllossom/llama-3-Korean-Bllossom-70B
library_name: transformers
tags:
  - mergekit
  - merge

🌋🌋 Huggingface Upload Issue

Maximum individual file size is 50.0GB to upload huggingface. To clear it, Split the file into part_aa, part_ab, part_ac chunks. After you download this repo on your folder path, command like this

cat part_* > llama-3-korean-bllossom-120b-Q4_K_M.gguf

Perhaps this will be the first GGUF model to upload such a large GGUF file of over 50GB to huggingface?

Sometimes we need to use a trick called chunks. 

Please wait to upload.....

🇰🇷 About the JayLee "AsianSoul"

"A leader who can make you rich 💵 !!!"

"Prove yourself with actual results, not just saying I know more than you!!!"

About this model

This is a 128B model based on Bllossom/llama-3-Korean-Bllossom-70B

☕ I started this Korean 120B model merge while drinking an iced Americano at Starbucks.

🏎️ I am a person whose goal is to turn the great results created by great genius scientists & groups around the world into profitable ones.

I hope this 120B is a helpful model for your future.

🌍 Collaboration is always welcome.🌍

About Bllossom/llama-3-Korean-Bllossom-70B
- Full model released in Korean over 100GB by Blossom team
- First in Korean! Expansion of Korean vocabulary to over 30,000 words
- Capable of processing Korean context that is approximately 25% longer than Llama3
- Connecting Korean-English knowledge using the Korean-English Parallel Corpus (pre-study)
- Fine tuning using data produced by linguists considering Korean culture and language
- Reinforcement learning

🛰️ About asiansoul/llama-3-Korean-Bllossom-120B-GGUF
- Q4_K_M : On a GPU with 68GB / more OR a CPU with 68G / more memory
- More Quantization ones i hope to upload, but your computer won't be able to handle it then. you know what i mean!!

Models Merged

The following models were included in the merge:

• Bllossom/llama-3-Korean-Bllossom-70B

Ollama

Create

ollama create Bllossom -f ./Modelfile_Q4_K_M 

MODELFILE (you can change this for your preference)

FROM llama-3-korean-bllossom-120b-Q4_K_M.gguf
TEMPLATE """{{ if .System }}<|start_header_id|>system<|end_header_id|>

{{ .System }}<|eot_id|>{{ end }}{{ if .Prompt }}<|start_header_id|>user<|end_header_id|>

{{ .Prompt }}<|eot_id|>{{ end }}<|start_header_id|>assistant<|end_header_id|>

{{ .Response }}<|eot_id|>"""


SYSTEM """
친절한 챗봇으로서 상대방의 요청에 최대한 자세하고 친절하게 답하자. 모든 대답은 한국어(Korean)으로 대답해줘.
"""

PARAMETER num_keep 24
PARAMETER temperature 0.7
PARAMETER num_predict 3000
PARAMETER stop "<|start_header_id|>"
PARAMETER stop "<|end_header_id|>"
PARAMETER stop "<|eot_id|>"

Configuration

The following YAML configuration was used to produce this model:

slices:
- sources:
  - layer_range: [0, 20]
    model: Bllossom/llama-3-Korean-Bllossom-70B
- sources:
  - layer_range: [10, 30]
    model: Bllossom/llama-3-Korean-Bllossom-70B
- sources:
  - layer_range: [20, 40]
    model: Bllossom/llama-3-Korean-Bllossom-70B
- sources:
  - layer_range: [30, 50]
    model: Bllossom/llama-3-Korean-Bllossom-70B
- sources:
  - layer_range: [40, 60]
    model: Bllossom/llama-3-Korean-Bllossom-70B
- sources:
  - layer_range: [50, 70]
    model: Bllossom/llama-3-Korean-Bllossom-70B
- sources:
  - layer_range: [60, 80]
    model: Bllossom/llama-3-Korean-Bllossom-70B
merge_method: passthrough
dtype: float16