Amigo_1.0

Amigo 1.0 - Coding Specialist

<div align="center"> <h3>AI-Powered Code Generation & Debugging</h3> <p><strong>Created by Jan Israel</strong></p> <p>Part of the Swordfish AI Trio</p> </div>

Model Details

Model Description

Amigo 1.0 is a specialized AI assistant fine-tuned for code generation and debugging. Built on CodeLlama-7B with LoRA adapters, Amigo excels at:

• Writing clean, efficient code in multiple programming languages
• Debugging complex codebases
• Explaining code logic and algorithms
• Providing code examples and implementations

Amigo means "Friend" in Spanish - your reliable coding companion.

• Developed by: Jan Francis Israel (The Swordfish)
• Model type: Causal Language Model with LoRA fine-tuning (PEFT)
• Language(s): English (code-focused)
• License: MIT
• Finetuned from: CodeLlama-7B-hf

Model Sources

• Repository: Part of the Swordfish Project
• Demo: Amigo-v2 Space | Amihan Space (Ensemble with Bandila)
• Sister Model: Bandila 1.0 (Reasoning Specialist)

⚠️ IMPORTANT: The Spaces are currently paused to manage cloud costs.

To test the live demo, please message @janisrael on HuggingFace to request Space activation.

The Spaces run on T4 Medium GPU ($0.60/hour) and are paused when not actively being showcased. When you request access, I'll resume them immediately. The first request takes 60-90 seconds (model loading), then normal speed afterward.

For more details about the pause/resume system and cost management, see the Amihan Space README.

Uses

Direct Use

Amigo 1.0 is designed for:

• Code Generation: Writing functions, classes, and complete programs
• Debugging Assistance: Identifying and fixing bugs in code
• Code Explanation: Understanding complex algorithms and patterns
• Programming Education: Learning best practices and patterns

Recommended Use with Amihan Ensemble

For best results, use Amigo alongside Bandila 1.0 through the Amihan Ensemble, which intelligently routes queries to the appropriate specialist.

Out-of-Scope Use

• Not suitable for: Production-critical code without human review
• Limitations: May generate code with bugs or security vulnerabilities
• Important: Always review and test generated code before deployment

How to Get Started

Installation

pip install transformers peft torch

Basic Usage

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel
import torch

# Load base model
base_model = AutoModelForCausalLM.from_pretrained(
    "codellama/CodeLlama-7b-hf",
    load_in_4bit=True,
    device_map="auto",
    torch_dtype=torch.float16
)

# Load Amigo LoRA adapter
model = PeftModel.from_pretrained(base_model, "swordfish7412/Amigo_1.0")
tokenizer = AutoTokenizer.from_pretrained("swordfish7412/Amigo_1.0")

# Generate code
prompt = "Instruction: Write a Python function to calculate factorial\nInput: \nOutput: "
inputs = tokenizer(prompt, return_tensors="pt").to("cuda")

outputs = model.generate(
    **inputs,
    max_new_tokens=200,
    temperature=0.7,
    do_sample=True
)

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

Training Details

Training Data

Amigo 1.0 was trained on:

• HumanEval (164 samples): Python code generation tasks
• Identity Dataset (495 samples): Custom identity and capability descriptions
• Total: 659 training samples across 3.62 epochs

Training Procedure

Training Configuration:

• Method: LoRA (Low-Rank Adaptation) fine-tuning with 4-bit quantization
• Base Model: codellama/CodeLlama-7b-hf
• Training Steps: 300
• Training Time: ~17 minutes on RTX A5000 (24GB)
• Hardware: RunPod Cloud GPU (RTX A5000)
• Framework: HuggingFace Transformers + PEFT

Hyperparameters:

• Batch Size: 2
• Gradient Accumulation: 4
• Learning Rate: 2e-4
• Max Length: 512 tokens
• LoRA Rank: 32
• LoRA Alpha: 64
• Optimizer: paged_adamw_8bit
• FP16: True

Training Results

• Initial Loss: 13.81
• Final Loss: 12.68
• Training Speed: 3.35s/step
• Model Size: 129MB (LoRA adapter only)

Identity & Capabilities

Amigo 1.0 knows its identity and purpose:

Name: Amigo 1.0
Creator: Jan Francis Israel (The Swordfish)
Role: Super Debugger AI - Coding Specialist
Specialties: Code generation, debugging, algorithm implementation

Evaluation

Testing Results

Amigo successfully generates:

• ✅ Python functions with proper docstrings
• ✅ Recursive algorithms (factorial, Fibonacci)
• ✅ Clean, readable code
• ✅ Correct identity responses

Example Outputs

Query: "Write a Python function to calculate factorial"

Amigo's Response:

def factorial(n):
    """
    This function returns the factorial of a given number.
    """
    return 1 if n == 0 else n * factorial(n - 1)

Bias, Risks, and Limitations

Known Limitations

• Code Quality: May generate inefficient or buggy code
• Security: Does not guarantee secure code practices
• Context: Limited to 512 tokens per query
• Languages: Primarily trained on Python, other languages may be less accurate

Recommendations

• Always review generated code
• Test code thoroughly before production use
• Use appropriate security scanning tools
• Consider this as a coding assistant, not a replacement for developers

Environmental Impact

• Hardware: RunPod RTX A5000 (24GB)
• Training Time: ~17 minutes
• Power Consumption: Minimal (single GPU, short training)
• Carbon Footprint: Negligible due to short training duration

Technical Specifications

Model Architecture

• Base: CodeLlama-7B (7 billion parameters)
• Adapter: LoRA with rank 32
• Quantization: 4-bit (nf4) via bitsandbytes
• Adapter Size: 129MB
• Total Parameters (with base): ~7B

Compute Infrastructure

• Provider: RunPod Cloud
• GPU: NVIDIA RTX A5000 (24GB VRAM)
• Training Framework: PyTorch + HuggingFace Transformers
• Quantization: bitsandbytes 4-bit

Filipino AI Squad 🇵🇭

Amigo is part of a trio of specialized AI models:

• 🔧 Amigo 1.0 (You are here) - Coding Specialist
• 🧠 Bandila 1.0 - Reasoning Specialist
• 🌬️ Amihan 1.0 - Intelligent Ensemble

Citation

@misc{amigo2024,
  author = {Jan Francis Israel},
  title = {Amigo 1.0: Super Debugger AI - Coding Specialist},
  year = {2024},
  publisher = {HuggingFace},
  howpublished = {\url{https://huggingface.co/swordfish7412/Amigo_1.0}},
  note = {Part of the Swordfish Project}
}

Model Card Authors

Jan Francis Israel (The Swordfish)

License

MIT License - Free to use with attribution

---

<div align="center"> <p><strong>Part of the Swordfish Project</strong></p> <p>Building elite AI debugging tools for developers worldwide 🇵🇭</p> </div>