113 lines
3.7 KiB
Markdown
113 lines
3.7 KiB
Markdown
---
|
||
library_name: transformers
|
||
tags:
|
||
- math
|
||
- code
|
||
- text-generation-inference
|
||
- llama3.2
|
||
license: apache-2.0
|
||
language:
|
||
- en
|
||
base_model:
|
||
- meta-llama/Llama-3.2-3B-Instruct
|
||
pipeline_tag: text-generation
|
||
---
|
||
|
||
|
||
# **Flerovium-Llama-3B**
|
||
|
||
> **Flerovium-Llama-3B** is a compact, general-purpose language model based on the powerful **llama 3.2** (llama) architecture. It is fine-tuned for a broad range of tasks including **mathematical reasoning**, **code generation**, and **natural language understanding**, making it a versatile choice for developers, students, and researchers seeking reliable performance in a lightweight model.
|
||
|
||
> \[!note]
|
||
> GGUF: [https://huggingface.co/prithivMLmods/Flerovium-Llama-3B-GGUF](https://huggingface.co/prithivMLmods/Flerovium-Llama-3B-GGUF)
|
||
|
||
---
|
||
|
||
## **Key Features**
|
||
|
||
1. **LLaMA 3.2 Backbone**
|
||
Built on **Meta’s LLaMA 3.2 (3B)** architecture, offering state-of-the-art performance in a compact footprint with better instruction-following and multilingual support.
|
||
|
||
2. **Multi-Task Fine-Tuning**
|
||
Finetuned on a modular and diverse dataset combining math, code, and general-purpose tasks—enabling clear explanations, problem solving, and practical utility.
|
||
|
||
3. **Strong Mathematical Reasoning**
|
||
Handles algebra, calculus, logic, and numerical problems with step-by-step clarity. Ideal for tutoring and academic use cases.
|
||
|
||
4. **Coding Capabilities**
|
||
Understands and generates clean, bug-free code in Python, JavaScript, C++, and more. Also excels at debugging, documentation, and logic explanations.
|
||
|
||
5. **General-Purpose Utility**
|
||
Performs well across everyday reasoning tasks—summarization, Q\&A, content drafting, and structured generation (Markdown, LaTeX, JSON).
|
||
|
||
6. **Efficient & Deployable**
|
||
With only 3 billion parameters, Flerovium-Llama-3B is resource-efficient and suitable for local deployment, offline tools, and edge AI setups.
|
||
|
||
---
|
||
|
||
## **Quickstart with Transformers**
|
||
|
||
```python
|
||
from transformers import AutoModelForCausalLM, AutoTokenizer
|
||
|
||
model_name = "prithivMLmods/Flerovium-Llama-3B"
|
||
|
||
model = AutoModelForCausalLM.from_pretrained(
|
||
model_name,
|
||
torch_dtype="auto",
|
||
device_map="auto"
|
||
)
|
||
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
||
|
||
prompt = "Explain how to solve a quadratic equation step-by-step."
|
||
|
||
messages = [
|
||
{"role": "system", "content": "You are a helpful AI assistant for math and coding."},
|
||
{"role": "user", "content": prompt}
|
||
]
|
||
|
||
text = tokenizer.apply_chat_template(
|
||
messages,
|
||
tokenize=False,
|
||
add_generation_prompt=True
|
||
)
|
||
|
||
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)
|
||
|
||
generated_ids = model.generate(
|
||
**model_inputs,
|
||
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]
|
||
print(response)
|
||
```
|
||
|
||
---
|
||
|
||
## **Intended Use**
|
||
|
||
* General-purpose text and reasoning
|
||
* Math tutoring and problem-solving
|
||
* Code generation, review, and debugging
|
||
* Content drafting in Markdown, LaTeX, and JSON
|
||
* Lightweight deployment in educational and developer environments
|
||
|
||
---
|
||
|
||
## **Limitations**
|
||
|
||
* Limited context length compared to large models (>7B)
|
||
* May require prompt refinement for very complex code/math problems
|
||
* Not ideal for long-form creative writing or deep conversational tasks
|
||
* Knowledge is limited to training data (no real-time web search)
|
||
|
||
---
|
||
|
||
## **References**
|
||
|
||
1. [LLaMA 3 Technical Report (Meta)](https://ai.meta.com/llama/)
|
||
2. [YaRN: Efficient Context Window Extension of Large Language Models](https://arxiv.org/pdf/2309.00071) |