---
library_name: peft
base_model: jangmin/midm-7b-safetensors-only
---

# Model Card for Model ID

<!-- Provide a quick summary of what the model is/does. -->
euneeei/hw-midm-7B-nsmc


### Training Data

<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
한국어로 된 네이버 영화 리뷰 데이터셋입니다.


#### Training Hyperparameters

- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
training_args: TrainingArguments = field(
        default_factory=lambda: TrainingArguments(
            output_dir="./results",
            max_steps=500,
            logging_steps=20,
            # save_steps=10,
            per_device_train_batch_size=1,
            per_device_eval_batch_size=1,
            gradient_accumulation_steps=2,
            gradient_checkpointing=False,
            group_by_length=False,
            # learning_rate=1e-4,
            learning_rate = 2e-4,
            lr_scheduler_type="cosine",
            warmup_steps=100,
            warmup_ratio=0.03,
            max_grad_norm=0.3,
            weight_decay=0.05,
            save_total_limit=20,
            save_strategy="epoch",
            num_train_epochs=1,
            optim="paged_adamw_32bit",
            fp16=True,
            remove_unused_columns=False,
            report_to="wandb",
        )
  
#### Speeds, Sizes, Times [optional]

<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->

[More Information Needed]

## Evaluation

<!-- This section describes the evaluation protocols and provides the results. -->

### Testing Data, Factors & Metrics

#### Testing Data

<!-- This should link to a Dataset Card if possible. -->
1000개

[More Information Needed]

#### Factors

<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
 learning_rate : 1e-4-> 2e-4
 max_steps=500 설정
 warmup_steps=100 설정

 
[More Information Needed]

#### Metrics

<!-- These are the evaluation metrics being used, ideally with a description of why. -->


precision	recall	f1-score	support
negative	0.87	0.95	091	492
positive	0.94	0.87	0.90	508
accuracy			0.91	1000
macro avg	0.91	0.91	0.91	1000
weighted avg	0.91	0.91	0.91	1000

confusion metrics

[[ 466, 26 ]
[68, 440]]

[More Information Needed]

### Results
정확도 0.51 -> 0.91로 높아졌습니다
[More Information Needed]


## Training procedure

The following `bitsandbytes` quantization config was used during training:
- quant_method: bitsandbytes
- load_in_8bit: False
- load_in_4bit: True
- llm_int8_threshold: 6.0
- llm_int8_skip_modules: None
- llm_int8_enable_fp32_cpu_offload: False
- llm_int8_has_fp16_weight: False
- bnb_4bit_quant_type: nf4
- bnb_4bit_use_double_quant: False
- bnb_4bit_compute_dtype: bfloat16

### Framework versions

- PEFT 0.7.0