---
tags:
- autotrain
- text-generation
- transformers
- named entity recognition
widget:
- text: 'I love AutoTrain because '
license: mit
datasets:
- conll2012_ontonotesv5
language:
- en
---


# Phi-2 model fine-tuned for named entity recognition task
The model was fine-tuned using one quarter of the ConLL 2012 OntoNotes v5 dataset.
- Dataset Source: [conll2012_ontonotesv5](https://huggingface.co/datasets/conll2012_ontonotesv5)
- Subset Used: English_v12
- Number of Examples: 87,265
  
The prompts and expected outputs were constructed as described in [1].

Example input:
```md
Instruct: I am an excelent linquist. The task is to label organization entities in the given sentence. Below are some examples

Input: A spokesman for B. A. T said of the amended filings that,`` It would appear that nothing substantive has changed.
Output: A spokesman for @@B. A. T## said of the amended filings that,`` It would appear that nothing substantive has changed.

Input: Since NBC's interest in the Qintex bid for MGM / UA was disclosed, Mr. Wright has n't been available for comment.
Output: Since @@NBC##'s interest in the @@Qintex## bid for @@MGM / UA## was disclosed, Mr. Wright has n't been available for comment.

Input: You know news organizations demand total transparency whether you're General Motors or United States government /.
Output: You know news organizations demand total transparency whether you're @@General Motors## or United States government /.

Input: We respectfully invite you to watch a special edition of Across China.
Output:
```
Expected output:
```md
We respectfully invite you to watch a special edition of @@Across China##.
```

This model is trained to recognize the named entity categories
- person
- nationalities or religious or political groups
- facility
- organization
- geopolitical entity
- location
- product
- date
- time expression
- percentage
- monetary value
- quantity
- event
- work of art
- law/legal reference
- language name

# Model Trained Using AutoTrain

This model was trained using **SFT** AutoTrain trainer. For more information, please visit [AutoTrain](https://hf.co/docs/autotrain).

Hyperparameters:
```json
{
    "model": "microsoft/phi-2",
    "valid_split": null,
    "add_eos_token": false,
    "block_size": 1024,
    "model_max_length": 1024,
    "padding": "right",
    "trainer": "sft",
    "use_flash_attention_2": false,
    "disable_gradient_checkpointing": false,
    "evaluation_strategy": "epoch",
    "save_total_limit": 1,
    "save_strategy": "epoch",
    "auto_find_batch_size": false,
    "mixed_precision": "bf16",
    "lr": 0.0002,
    "epochs": 1,
    "batch_size": 1,
    "warmup_ratio": 0.1,
    "gradient_accumulation": 4,
    "optimizer": "adamw_torch",
    "scheduler": "linear",
    "weight_decay": 0.01,
    "max_grad_norm": 1.0,
    "seed": 42,
    "apply_chat_template": false,
    "quantization": "int4",
    "target_modules": null,
    "merge_adapter": false,
    "peft": true,
    "lora_r": 16,
    "lora_alpha": 32,
    "lora_dropout": 0.05,
    "dpo_beta": 0.1,
}
```

# Usage

```python

from transformers import AutoModelForCausalLM, AutoTokenizer

model_path = "pahautelman/phi2-ner-v1"

tokenizer = AutoTokenizer.from_pretrained(model_path)
model = AutoModelForCausalLM.from_pretrained(
    model_path
).eval()

prompt = 'Label the person entities in the given sentence: Russian President Vladimir Putin is due to arrive in Havana a few hours from now to become the first post-Soviet leader to visit Cuba.'

inputs = tokenizer.encode(prompt, add_special_tokens=False, return_tensors='pt')
outputs = model.generate(
    inputs.to(model.device),
    max_new_tokens=9,
    do_sample=False,
)
output = tokenizer.batch_decode(outputs)[0]

# Model response: "Output: Russian President, Vladimir Putin"
print(output)
```

# References:
[1] Wang et al., GPT-NER: Named entity recognition via large language models 2023