103 lines
3.9 KiB
Markdown
103 lines
3.9 KiB
Markdown
---
|
|
library_name: transformers
|
|
tags:
|
|
- text-generation-inference
|
|
- pretraining/SFT
|
|
- code
|
|
- math
|
|
license: apache-2.0
|
|
language:
|
|
- en
|
|
base_model:
|
|
- Gensyn/Qwen2.5-0.5B-Instruct
|
|
pipeline_tag: text-generation
|
|
---
|
|
|
|
|
|
|
|
# **Lang-Exster-0.5B-Instruct**
|
|
|
|
> **Lang-Exster-0.5B-Instruct** is a **general-purpose instruction-following LLM** fine-tuned from **Qwen2.5-0.5B**. This model is optimized for **lightweight deployments** and **instructional clarity**, capable of performing a wide range of natural language and programming-related tasks with efficiency and interpretability.
|
|
|
|
## **Key Features**
|
|
|
|
1. **Instruction Following & Explanation**
|
|
Trained to **understand, follow, and respond** to natural language instructions with clear, logical, and relevant output. Suitable for Q&A, step-by-step reasoning, and guided code generation.
|
|
|
|
2. **Lightweight General-Purpose Model**
|
|
Fine-tuned from **Qwen2.5-0.5B**, making it **highly efficient for edge devices**, **local tools**, and **low-resource applications** without sacrificing utility.
|
|
|
|
3. **Multi-Domain Task Handling**
|
|
Can perform across **coding**, **writing**, **summarization**, **chat**, **translation**, and **educational queries**, thanks to its broad general-purpose instruction tuning.
|
|
|
|
4. **Compact and Efficient**
|
|
At just **0.5B parameters**, Lang-Exster is optimized for **fast inference**, **low memory usage**, and seamless integration into developer tools and workflows.
|
|
|
|
5. **Code Assistance (Lite)**
|
|
Capable of **basic code generation**, **syntax checking**, and **conceptual explanations**, especially useful for beginners and instructional applications.
|
|
|
|
## **Quickstart with Transformers**
|
|
|
|
```python
|
|
from transformers import AutoModelForCausalLM, AutoTokenizer
|
|
|
|
model_name = "prithivMLmods/Lang-Exster-0.5B-Instruct"
|
|
|
|
model = AutoModelForCausalLM.from_pretrained(
|
|
model_name,
|
|
torch_dtype="auto",
|
|
device_map="auto"
|
|
)
|
|
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
|
|
|
prompt = "Write a Python function that checks if a number is prime, and explain how it works."
|
|
|
|
messages = [
|
|
{"role": "system", "content": "You are an instructional assistant. Follow user instructions clearly and explain your reasoning."},
|
|
{"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]
|
|
```
|
|
|
|
## **Intended Use**
|
|
|
|
- **General-Purpose Assistant**:
|
|
Performs everyday tasks such as Q&A, summarization, light coding, language generation, and translation.
|
|
|
|
- **Educational Support**:
|
|
Aids learners in understanding topics through **guided explanations**, **basic coding help**, and **concept breakdowns**.
|
|
|
|
- **Lightweight Developer Integration**:
|
|
Ideal for command-line assistants, browser plugins, and desktop utilities with limited compute resources.
|
|
|
|
- **Instruction Clarity Demonstrator**:
|
|
Acts as a fine baseline for developing **instruction-tuned** capabilities in constrained environments.
|
|
|
|
## **Limitations**
|
|
|
|
1. **Scale Limitations**
|
|
Being a 0.5B model, it has limited memory and may not handle deep context or long documents effectively.
|
|
|
|
2. **Reasoning Depth**
|
|
Provides **surface-level reasoning** and may struggle with highly technical, abstract, or creative prompts.
|
|
|
|
3. **Basic Code Generation**
|
|
Supports basic scripting and logic but may miss edge cases or advanced patterns in complex code.
|
|
|
|
4. **Prompt Design Sensitivity**
|
|
Performs best with **clear**, **concise**, and **well-structured** instructions. |