README.md

# Job Tag Embedding Model (Dev)

Fine-tuned embedding model for job category recommendation based on [BAAI/bge-large-zh-v1.5](https://huggingface.co/BAAI/bge-large-zh-v1.5).

## Model Details

- **Base Model:** 1111DataScience/job_tag_embedding
- **Training Data:** Job titles and category pairs
- **Training Steps:** 1,920 (3 epochs)
- **Final Loss:** 2.126

## Usage

```python
from FlagEmbedding import FlagModel

# Load model
model = FlagModel('1111DataScience/job_tag_embedding_dev', use_fp16=True)

# Encode query (job title)
query_embedding = model.encode_queries(["內外場儲備幹部"])

# Encode candidates (job categories)
candidate_embeddings = model.encode([
    "儲備幹部",
    "餐廚助手",
    "餐飲服務人員"
])

# Calculate similarity
similarities = query_embedding @ candidate_embeddings.T
```

## Training Command

```bash
torchrun --nproc_per_node 1 \
    -m FlagEmbedding.finetune.embedder.encoder_only.base \
    --model_name_or_path 1111DataScience/job_tag_embedding \
    --cache_dir ./cache/model \
    --train_data training_data.jsonl \
    --output_dir ./output
```