Compact Dhivehi (ދިވެހި) text generation model based on google/gemma-3-270m, designed to generate creative and coherent Dhivehi content based on prompts, titles, or instructions.

Note: This model is specifically tuned for text generation tasks and provides natural, flowing Dhivehi text outputs for content creation.

Model details

Base: google/gemma-3-270m-it
Language: Dhivehi
Style: Creative, natural, coherent
Output format: Flowing Dhivehi text
Supported tasks:
- Random Dhivehi text generation

Intended use

To further downstream tasks

How to use

python

from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
import torch

# Load model
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
import torch

# Load model
model_path = "alakxender/gemma-3-270m-dhivehi-text-gen"
model = AutoModelForCausalLM.from_pretrained(
    model_path, 
    torch_dtype="auto", 
    device_map="auto", 
    attn_implementation="eager"
)
tokenizer = AutoTokenizer.from_pretrained(model_path)
pipe = pipeline("text-generation", model=model, tokenizer=tokenizer)

# Generate content
title_or_prompt = "ދިވެހިރާއްޖެއަކީ އިންޑިޔާ ކަނޑުގައި އޮންނަ ޖަޒީރާ ޤައުމެކެވެ"

# Create the prompt format used during training
prompt = f"Create a dhivehi article for the following topic: {title_or_prompt}"

# Create chat format message for content generation (matching training format)
messages = [
    {"role": "system", "content": "You are a helpful assistant that can generate dhivehi articles based on a given topic."},
    {"role": "user", "content": prompt}
]

# Apply chat template
formatted_prompt = pipe.tokenizer.apply_chat_template(
    messages, 
    tokenize=False, 
    add_generation_prompt=True
)

# Generation parameters
gen_kwargs = {
    "max_new_tokens": 256,
    "temperature": 0.7,
    "top_p": 0.9,
    "top_k": 50,
    "do_sample": True,
    "disable_compile": True,
    "pad_token_id": tokenizer.eos_token_id
}

# Generate content
outputs = pipe(formatted_prompt, **gen_kwargs)

# Extract generated content (remove the prompt)
generated_content = outputs[0]['generated_text'][len(formatted_prompt):].strip()

print(f"Generated content: {generated_content}")

Generation Parameters

max_new_tokens: Controls the length of generated text (64-512 recommended)
temperature: Controls randomness (0.1-1.0, higher = more creative)
do_sample: Boolean flag to enable/disable sampling

Limitations

Generated content may not always be factually accurate
Quality depends on the clarity and specificity of input prompts
Context window limitations for very long inputs
The model is specifically trained for text generation tasks in Dhivehi

alakxender/gemma-3-270m-dhivehi-content-gen

Gemma-3-270M Dhivehi — Text Generation Model

Model details

Intended use

How to use

Generation Parameters

Limitations

alakxender/
gemma-3-270m-dhivehi-content-gen