README.md

---
license: apache-2.0
base_model:
- mistralai/Mistral-Nemo-Base-2407
language:
- en
- ko
- ja
- zh
datasets:
- 4DR1455/finance_questions
- Aratako/Synthetic-JP-Conversations-Magpie-Nemotron-4-10k
- Aratako/Synthetic-JP-EN-Coding-Dataset-Magpie-69k
- Aratako/Synthetic-Japanese-Roleplay-NSFW-Claude-3.5s-10.5k-formatted
- BCCard/BCCard-Finance-Kor-QnA
- CarrotAI/ko-code-alpaca-QA
- ChuGyouk/AI_healthcare_QA_samples_Sonnet3.5
- DavidLanz/medical_instruction
- Dusker/lawyer-llama
- Gryphe/Sonnet3.5-Charcard-Roleplay
- HAERAE-HUB/qarv-instruct-ko
- HachiML/alpaca_jp_math
- Magpie-Align/Magpie-Llama-3.1-Pro-MT-300K-v0.1
- Magpie-Align/Magpie-Qwen2-Pro-200K-Chinese
- beomi/KoAlpaca-v1.1a
- codefuse-ai/Evol-instruction-66k
- frankminors123/belle-math-zh
- gbharti/wealth-alpaca_lora
- iam-ajaymeena/Self-Instruct-Japanese-Elzya-13B
- jihye-moon/LawQA-Ko
- jondurbin/gutenberg-dpo-v0.1
- junyeong-nero/kin_med_100K_edited
- kyujinpy/KOR-OpenOrca-Platypus-v3
- lavita/medical-qa-datasets
- microsoft/orca-math-word-problems-200k
- neural-bridge/rag-dataset-12000
- p1atdev/ichikara-instruction
- qiaojin/PubMedQA
- shibing624/roleplay-zh-sharegpt-gpt4-data
- team-hatakeyama-phase2/AutoMultiTurnByCalm3-22B-Corrected-reformatted
- ymoslem/Law-StackExchange
- zzunyang/LawQA_LawSee
---
# Mistral-Nemo-NT-Ko-12B-sft

## Description

**Mistral-Nemo-NT-Ko-12B-sft** is an instruction-tuned version of [*mistralai/Mistral-Nemo-Base-2407*](https://huggingface.co/mistralai/Mistral-Nemo-Base-2407), fine-tuned across four languages: English, Korean, Chinese, and Japanese.

The primary goals of this model are **language alignment**, **cross-lingual knowledge transfer** and **ChatML formatting**. This is an intermediate version since preference optimization has not yet been applied.


## Features

- The base model supports a context length of 128K, while I fine-tuned this model with an 8K context size.

- The model follows to the input language unless the user explicitly specifies an output language (If the language is set by a system role, it may be ignored).

- Answer length tends to vary by language: English responses are generally longer than average, while Korean responses tend to be shorter. The behavior for Japanese and Chinese is still under observation.

- Recommended temperature settings: 0.3 to 0.7.


# Evaluation

## LogicKor

| 모델 | 방법 | 추론 | 수학 | 글쓰기 | 코딩 | 이해 | 문법 | 싱글턴 | 멀티턴 | 총점 |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
|Mistral-Nemo-NT-Ko-12B-sft| cot-1-shot |7.36 | 6.57 | 8.71 | 8.57 | 9.57 | 6.43 | 7.81 | 7.93 | **7.87** |
|Mistral-Nemo-NT-Ko-12B-sft| 1-shot | 9.00 | 5.71 | 7.93 | 8.29 | 7.93 | 5.21 | 7.29 | 7.40 | 7.35 |
| Mistral Nemo | 1-shot | 5.00, | 6.50 | 6.86 | 8.07 | 7.64 | 8.43 | 7.60 | 6.57 |7.08|
| Mistral Nemo | cot-1-shot | 5.43, | 6.86 | 6.07 | 7.57 | 5.86 | 7.57 | 7.50 | 5.62 |6.56|
|Mistral-Nemo-NT-Ko-12B-sft| default | 6.00 | 4.93 | 5.43 | 7.14 | 9.71 | 4.00 | 6.45 | 5.95 | 6.20 |
| Mistral Nemo | default | 0.43, | 7.64 | 6.21 | 7.14 | 6.79 | 7.21 | 6.26 | 5.55 |5.90|

## MT-Bench

| Model | First | Second | Average |
| --- | --- | --- | --- |
|Mistral-Nemo-NT-Ko-12B-sft| 8.39 | 7.99 | 8.19 |
\* ```judge-model: GPT-4```

## Language-Confusion(Korean Only)

| Model | Monolingual-LPR | Monolingual-WPR | Crosslingual-LPR | Crosslingual-WPR |
| --- | --- | --- | --- | --- |
|Mistral-Nemo-NT-Ko-12B-sft| 100.00% | 99.00% | 87.51% | 96.96% |
|Mistral-Nemo-Instruct-2407 | 90.72% | 93.18% | 46.75% | 92.84% |
|Meta-Llama-3.1-8B-Instruct | 99.00% | 96.97% | 91.45% | 93.01% |
|gemma-2-9b-it | 100.00% | 98.00% | 87.93% | 95.58% |


example:

```
<|im_start|>system
You are a helpful AI assistant.<|im_end|>
<|im_start|>user
{prompt}<|im_end|>
<|im_start|>assistant
```

*I trained Mistral-Nemo-NT-Ko-12B with various system prompt from dozens of dataset. You can chat with/without your system prompt.*


# Dataset

[werty1248/multilingual-instruct-balanced](https://huggingface.co/datasets/werty1248/multilingual-instruct-balanced)

# Training Details

- GPU: 8xA40
- epoch: 3
- total batch size: 8
- learning rate: 7e-6
- weight decay: 0.01

[<img src="https://raw.githubusercontent.com/axolotl-ai-cloud/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>](https://github.com/axolotl-ai-cloud/axolotl)
<details><summary>See axolotl config</summary>

axolotl version: `0.4.1`
```yaml
base_model: mistralai/Mistral-Nemo-Base-2407
model_type: MistralForCausalLM
tokenizer_config: nothingiisreal/MN-12B-Celeste-V1.9 ##axolotl-ai-co/Mistral-Nemo-Base-2407-chatml makes error, why?
tokenizer_type: AutoTokenizer

load_in_8bit: false
load_in_4bit: false
strict: false

chat_template: chatml
datasets:
  - path: werty1248/multilingual-instruct-balanced
    type: sharegpt
    chat_template: chatml

dataset_prepared_path: ./data_preparation
output_dir: /workspace/data

hf_use_auth_token: true

sequence_len: 8192
sample_packing: true
pad_to_sequence_len: true

wandb_project:
#wandb_entity:
#wandb_watch:
wandb_name:
#wandb_log_model:

gradient_accumulation_steps: 1 ## total_batch = 8
micro_batch_size: 1
num_epochs: 3
optimizer: paged_adamw_32bit
lr_scheduler: cosine
learning_rate: 0.000007

train_on_inputs: false
group_by_length: false
bf16: auto
fp16: 
tf32: false

gradient_checkpointing: true
early_stopping_patience:
resume_from_checkpoint:
local_rank:
logging_steps: 1
xformers_attention:
flash_attention: true

warmup_steps: 1000
evals_per_epoch: 1
eval_table_size:
save_steps: 1000
debug:
deepspeed: deepspeed_configs/zero3_bf16.json
weight_decay: 0.01
special_tokens:
  pad_token: <pad>
```

</details><br>


- Training loss

![image/png](https://cdn-uploads.huggingface.co/production/uploads/6629154d55d7c289634b8c5d/Xcat10ejYX1nU4cH94vJF.png)