Instructions to use devshaheen/llama-3.2-3b-Instruct-finetune with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use devshaheen/llama-3.2-3b-Instruct-finetune with Transformers:

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

pipe = pipeline("text-generation", model="devshaheen/llama-3.2-3b-Instruct-finetune")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("devshaheen/llama-3.2-3b-Instruct-finetune")
model = AutoModelForCausalLM.from_pretrained("devshaheen/llama-3.2-3b-Instruct-finetune", device_map="auto")
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 Settings

vLLM

How to use devshaheen/llama-3.2-3b-Instruct-finetune with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "devshaheen/llama-3.2-3b-Instruct-finetune"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "devshaheen/llama-3.2-3b-Instruct-finetune",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/devshaheen/llama-3.2-3b-Instruct-finetune

SGLang

How to use devshaheen/llama-3.2-3b-Instruct-finetune 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 "devshaheen/llama-3.2-3b-Instruct-finetune" \
    --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": "devshaheen/llama-3.2-3b-Instruct-finetune",
		"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 "devshaheen/llama-3.2-3b-Instruct-finetune" \
        --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": "devshaheen/llama-3.2-3b-Instruct-finetune",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Unsloth Studio

How to use devshaheen/llama-3.2-3b-Instruct-finetune with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for devshaheen/llama-3.2-3b-Instruct-finetune to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for devshaheen/llama-3.2-3b-Instruct-finetune to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for devshaheen/llama-3.2-3b-Instruct-finetune to start chatting

Load model with FastModel

pip install unsloth
from unsloth import FastModel
model, tokenizer = FastModel.from_pretrained(
    model_name="devshaheen/llama-3.2-3b-Instruct-finetune",
    max_seq_length=2048,
)

Docker Model Runner
How to use devshaheen/llama-3.2-3b-Instruct-finetune with Docker Model Runner:
```
docker model run hf.co/devshaheen/llama-3.2-3b-Instruct-finetune
```

devshaheen commited on Dec 1, 2024

Commit

ff0c9e9

verified ·

1 Parent(s): fb9c721

updated readme

Browse files

Files changed (1) hide show

README.md +75 -5

README.md CHANGED Viewed

@@ -6,17 +6,87 @@ tags:
 - unsloth
 - llama
 - trl
 license: apache-2.0
 language:
 - en
 ---
-# Uploaded  model
 - **Developed by:** devshaheen
-- **License:** apache-2.0
-- **Finetuned from model :** unsloth/llama-3.2-3b-instruct-bnb-4bit
-This llama model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library.
-[<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)

 - unsloth
 - llama
 - trl
+- multilingual
+- instruction-tuning
 license: apache-2.0
 language:
 - en
+- kn
+datasets:
+- charanhu/kannada-instruct-dataset-390k
+library_name: transformers
 ---
+# Uploaded Model: devshaheen/llama-3.2-3b-Instruct-finetune
+## Overview
 - **Developed by:** devshaheen
+- **License:** Apache-2.0
+- **Finetuned from model:** `unsloth/llama-3.2-3b-instruct-bnb-4bit`
+- **Languages Supported:**
+  - **English** (`en`) for general-purpose text generation and instruction-following tasks.
+  - **Kannada** (`kn`) with a focus on localized and culturally aware text generation.
+- **Dataset Used:** [charanhu/kannada-instruct-dataset-390k](https://huggingface.co/datasets/charanhu/kannada-instruct-dataset-390k)
+This model is a fine-tuned version of LLaMA, optimized for multilingual instruction-following tasks with a specific emphasis on English and Kannada. It utilizes 4-bit quantization for efficient deployment in low-resource environments without compromising performance.
+---
+## Features
+### 1. **Instruction Tuning**
+The model is trained to follow a wide range of instructions and generate contextually relevant responses. It excels in both creative and factual text generation tasks.
+### 2. **Multilingual Support**
+The model is capable of generating text in Kannada and English, making it suitable for users requiring bilingual capabilities.
+### 3. **Optimized Training**
+Training was accelerated using [Unsloth](https://github.com/unslothai/unsloth), achieving **2x faster training** compared to conventional methods. This was complemented by HuggingFace's TRL (Transformers Reinforcement Learning) library to ensure high performance.
+### 4. **Efficiency through Quantization**
+Built on the `bnb-4bit` quantized model, it is designed for optimal performance in environments with limited computational resources while maintaining precision and depth in output.
+---
+## Usage Scenarios
+### General Use
+- Text completion and creative writing.
+- Generating instructions or following queries in English and Kannada.
+### Specialized Applications
+- Localized AI systems in Kannada for chatbots, educational tools, and more.
+- Research and development on multilingual instruction-tuned models.
+---
+## Performance and Metrics
+### Evaluation Dataset:
+The model was fine-tuned on [charanhu/kannada-instruct-dataset-390k](https://huggingface.co/datasets/charanhu/kannada-instruct-dataset-390k), a comprehensive dataset designed for Kannada instruction tuning.
+### Training Parameters:
+- **Base Model:** LLaMA 3.2-3B-Instruct
+- **Optimizer:** AdamW
+- **Quantization:** 4-bit (bnb-4bit)
+- **Framework:** HuggingFace Transformers + Unsloth
+---
+## Example Usage
+### Python Code
+```python
+from transformers import AutoModelForCausalLM, AutoTokenizer
+# Load model and tokenizer
+model_name = "devshaheen/llama-3.2-3b-Instruct-finetune"
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+model = AutoModelForCausalLM.from_pretrained(model_name)
+# Generate text
+input_text = "How does climate change affect the monsoon in Karnataka?"
+inputs = tokenizer(input_text, return_tensors="pt")
+outputs = model.generate(**inputs, max_length=150)
+print(tokenizer.decode(outputs[0], skip_special_tokens=True))