Instructions to use prithivMLmods/Llama-Song-Stream-3B-Instruct with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use prithivMLmods/Llama-Song-Stream-3B-Instruct with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="prithivMLmods/Llama-Song-Stream-3B-Instruct")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("prithivMLmods/Llama-Song-Stream-3B-Instruct")
model = AutoModelForCausalLM.from_pretrained("prithivMLmods/Llama-Song-Stream-3B-Instruct")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use prithivMLmods/Llama-Song-Stream-3B-Instruct with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "prithivMLmods/Llama-Song-Stream-3B-Instruct"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "prithivMLmods/Llama-Song-Stream-3B-Instruct",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/prithivMLmods/Llama-Song-Stream-3B-Instruct

SGLang

How to use prithivMLmods/Llama-Song-Stream-3B-Instruct with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "prithivMLmods/Llama-Song-Stream-3B-Instruct" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "prithivMLmods/Llama-Song-Stream-3B-Instruct",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "prithivMLmods/Llama-Song-Stream-3B-Instruct" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "prithivMLmods/Llama-Song-Stream-3B-Instruct",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use prithivMLmods/Llama-Song-Stream-3B-Instruct with Docker Model Runner:
```
docker model run hf.co/prithivMLmods/Llama-Song-Stream-3B-Instruct
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

Llama-Song-Stream-3B-Instruct Model Card

The Llama-Song-Stream-3B-Instruct is a fine-tuned language model specializing in generating music-related text, such as song lyrics, compositions, and musical thoughts. Built upon the meta-llama/Llama-3.2-3B-Instruct base, it has been trained with a custom dataset focused on song lyrics and music compositions to produce context-aware, creative, and stylized music output.

File Name	Size	Description
`.gitattributes`	1.57 kB	LFS tracking file to manage large model files.
`README.md`	282 Bytes	Documentation with model details and usage.
`config.json`	1.03 kB	Model configuration settings.
`generation_config.json`	248 Bytes	Generation parameters like max sequence length.
`pytorch_model-00001-of-00002.bin`	4.97 GB	Primary weights (part 1 of 2).
`pytorch_model-00002-of-00002.bin`	1.46 GB	Primary weights (part 2 of 2).
`pytorch_model.bin.index.json`	21.2 kB	Index file mapping the checkpoint layers.
`special_tokens_map.json`	477 Bytes	Defines special tokens for tokenization.
`tokenizer.json`	17.2 MB	Tokenizer data for text generation.
`tokenizer_config.json`	57.4 kB	Configuration settings for tokenization.

Key Features

Song Generation:
- Generates full song lyrics based on user input, maintaining rhyme, meter, and thematic consistency.
Music Context Understanding:
- Trained on lyrics and song patterns to mimic and generate song-like content.
Fine-tuned Creativity:
- Fine-tuned using Song-Catalogue-Long-Thought for coherent lyric generation over extended prompts.
Interactive Text Generation:
- Designed for use cases like generating lyrical ideas, creating drafts for songwriters, or exploring themes musically.

Training Details

Base Model: meta-llama/Llama-3.2-3B-Instruct
Finetuning Dataset: prithivMLmods/Song-Catalogue-Long-Thought
- This dataset comprises 57.7k examples of lyrical patterns, song fragments, and themes.

Applications

Songwriting AI Tools:
- Generate lyrics for genres like pop, rock, rap, classical, and others.
Creative Writing Assistance:
- Assist songwriters by suggesting lyric variations and song drafts.
Storytelling via Music:
- Create song narratives using custom themes and moods.
Entertainment AI Integration:
- Build virtual musicians or interactive lyric-based content generators.

Example Usage

Setup

First, load the Llama-Song-Stream model:

from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "prithivMLmods/Llama-Song-Stream-3B-Instruct"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name)

Generate Lyrics Example

prompt = "Write a song about freedom and the open sky"
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_length=100, temperature=0.7, num_return_sequences=1)

generated_lyrics = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(generated_lyrics)

Deployment Notes

Serverless vs. Dedicated Endpoints:
The model currently does not have enough usage for a serverless endpoint. Options include:
- Dedicated inference endpoints for faster responses.
- Custom integrations via Hugging Face inference tools.
Resource Requirements:
Ensure sufficient GPU memory and compute for large PyTorch model weights.