slm_finetuned_T1

asrith05/slm_finetuned_T1

This is a fine-tuned multilingual model specialized for entity extraction tasks, based on DeepSeek architecture.

Model Details

• Base Model: DeepSeek multilingual pretrained model
• Languages: English, Telugu, Sanskrit
• Task: Entity extraction and named entity recognition
• Fine-tuning: Specialized for entity extraction from structured data
• Size: ~1251MB

Description

This model has been fine-tuned specifically for entity extraction tasks across multiple languages. It can identify and extract various types of entities from text, including:

• Person names
• Locations
• Organizations
• Products
• Events
• Dates and times
• Custom domain-specific entities

Usage

from transformers import AutoTokenizer, AutoModelForCausalLM

model_name = "asrith05/slm_finetuned_T1"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name)

# Example entity extraction
prompt = "Extract entities from: John Smith works at Microsoft in Seattle and attended the AI Conference."
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(
    **inputs, 
    max_new_tokens=100,
    temperature=0.3,
    do_sample=True
)
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(response)

Training Details

• Base Model: Multilingual DeepSeek pretrained model
• Fine-tuning Dataset: 20K entity extraction examples
• Validation Set: 5K examples
• Test Set: 5K examples
• Training: 1 epoch with validation monitoring
• Learning Rate: 5e-5
• Batch Size: 16 (effective)

Performance

The model demonstrates strong performance on:

• Multilingual entity recognition: Works across English, Telugu, and Sanskrit
• Cross-domain extraction: Handles various domains and entity types
• Structured data processing: Can process and extract from formatted data

Training Data Format

The model was trained on data in the format:

Input: "Extract entities from: [text]"
Output: "Entities: [extracted entities with labels]"

Limitations

• Optimized for entity extraction tasks
• May require prompt formatting for best results
• Performance may vary across different domains
• Generated content should be validated for accuracy

Related Models

• Base Model: asrith05/deepseek_pretrain_90k
• General Fine-tuned: asrith05/slm

License

This model is released under the Apache 2.0 License.

Citation

If you use this model, please cite:

@model{asrith05_slm_finetuned_T1,
  author = {Asrith},
  title = {Multilingual Entity Extraction Model},
  year = {2025},
  publisher = {Hugging Face},
  url = {https://huggingface.co/asrith05/slm_finetuned_T1}
}