---
language:
- ja
library_name: transformers
pipeline_tag: text-generation
---

# Model Card for Model ID

This model can be used to convert Japanese IPA back to normal text.




## Model 

### Usage

For vLLM use this: Respair/Test_QwJP


in the terminal:
```bash
python -m vllm.entrypoints.openai.api_server --model Respair/Japanese_Phoneme_to_Grapheme_LLM--port 8000
```

now you can simply use it:
```python

# pip install vllm

from openai import OpenAI


openai_api_key = "EMPTY"
openai_api_base = "http://localhost:8000/v1"

client = OpenAI(
    api_key=openai_api_key,
    base_url=openai_api_base,
)

model_name = "Respair/Test_QwJP"


def p2g(param):

    chat_response = client.chat.completions.create(

        model=model_name,
        max_tokens=512,
        temperature=0.01,

        messages=[
            
            {"role": "user", "content": f"{prompt}"}]
    )   

    return chat_response.choices[0].message.content


prompt= f""" Turn IPA to Japanese: geɴ'iɴ?  sonna  fɯɯ ni ɕiɽoi hebi no geŋkakɯ ga, omae no ɕɯɯi ni naɴ do mo naɴ do mo aɽawaɽerɯ, kiʔkake na no ka naɴ na no ka? mi ni oboeʔtsɯ no ka? """

result= p2g(prompt)

print(result)
```


...or simply through HF transformers:

```python
from transformers import AutoModelForCausalLM, AutoTokenizer
device = "cuda" # the device to load the model onto

model = AutoModelForCausalLM.from_pretrained(
    "Respair/Japanese_Phoneme_to_Grapheme_LLM",
    torch_dtype="auto",
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained("Respair/Japanese_Phoneme_to_Grapheme_LLM")


tokenizer.pad_token = "<|endoftext|>"
tokenizer.bos_token = "<|endoftext|>"
tokenizer.eos_token = "<|im_end|>"
prompt = "Turn IPA to Japanese: geɴ'iɴ?  sonna  fɯɯ ni ɕiɽoi hebi no geŋkakɯ ga, omae no ɕɯɯi ni naɴ do mo naɴ do mo aɽawaɽerɯ, kiʔkake na no ka naɴ na no ka? mi ni oboeʔtsɯ no ka?"
messages = [

    {"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True
)
model_inputs = tokenizer([text], return_tensors="pt").to(device)

generated_ids = model.generate(
    model_inputs.input_ids,

    pad_token_id=tokenizer.pad_token_id,
    bos_token_id=tokenizer.bos_token_id,
    eos_token_id=tokenizer.eos_token_id,

    temperature=0.1,
    max_new_tokens=512
)
generated_ids = [
    output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
]

response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
response
```