README.md

metadata

base_model: meta-llama/Llama-3.2-3B-Instruct
library_name: peft
pipeline_tag: text-generation
tags:
  - base_model:adapter:meta-llama/Llama-3.2-3B-Instruct
  - lora
  - transformers
  - configuration-management
  - secrets-management
  - devops
  - multi-cloud
license: mit
language:
  - en

AnySecret Assistant

A specialized AI assistant for AnySecret configuration management, fine-tuned on Llama-3.2-3B-Instruct to help with multi-cloud secrets and parameters management.

Model Details

Model Description

This is a LoRA fine-tuned version of Llama-3.2-3B-Instruct, specifically trained to assist with AnySecret configuration management across AWS, GCP, Azure, and Kubernetes environments. The model can help with CLI commands, configuration setup, CI/CD integration, and Python SDK usage.

• Developed by: anysecret-io
• Model type: Causal Language Model (LoRA Adapter)
• Language(s) (NLP): English
• License: MIT
• Finetuned from model: meta-llama/Llama-3.2-3B-Instruct

Model Sources

• Repository: https://github.com/anysecret-io/anysecret-lib
• Documentation: https://docs.anysecret.io
• Demo: Coming soon

Uses

Direct Use

This model is designed to provide expert assistance with:

• AnySecret CLI commands and usage patterns
• Multi-cloud configuration (AWS, GCP, Azure, Kubernetes)
• Secrets vs parameters classification and management
• CI/CD pipeline integration
• Python SDK implementation guidance

Example Usage

from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer

base_model = AutoModelForCausalLM.from_pretrained("meta-llama/Llama-3.2-3B-Instruct")
model = PeftModel.from_pretrained(base_model, "anysecret-io/anysecret-assistant")
tokenizer = AutoTokenizer.from_pretrained("meta-llama/Llama-3.2-3B-Instruct")

prompt = "### Instruction:\nHow do I configure AnySecret for AWS?\n\n### Response:\n"
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=256)

Out-of-Scope Use

This model is specifically trained for AnySecret configuration management and may not perform well on:

• General programming questions unrelated to configuration management
• Other secrets management tools or platforms
• Security vulnerabilities or exploitation techniques

Training Details

Training Data

The model was trained on 43 curated examples across 7 categories:

• CLI Commands (9 examples) - Command usage patterns
• AWS Configuration (6 examples) - AWS Secrets Manager integration
• GCP Configuration (6 examples) - Google Secret Manager setup
• Azure Configuration (6 examples) - Azure Key Vault integration
• Kubernetes (6 examples) - K8s secrets and ConfigMaps
• CI/CD Integration (5 examples) - GitHub Actions, Jenkins workflows
• Python Integration (5 examples) - SDK usage patterns

Training Procedure

Training Hyperparameters

• Base Model: meta-llama/Llama-3.2-3B-Instruct
• LoRA Rank: 16
• LoRA Alpha: 32
• Target Modules: q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj
• Learning Rate: 2e-4
• Batch Size: 1 (with gradient accumulation)
• Epochs: 2-3
• Training regime: fp16 mixed precision with 4-bit quantization

How to Get Started with the Model

# Install requirements
pip install torch transformers peft

# Load and use the model
from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

base_model = AutoModelForCausalLM.from_pretrained(
    "meta-llama/Llama-3.2-3B-Instruct",
    torch_dtype=torch.float16,
    device_map="auto"
)

model = PeftModel.from_pretrained(base_model, "anysecret-io/anysecret-assistant")
tokenizer = AutoTokenizer.from_pretrained("meta-llama/Llama-3.2-3B-Instruct")

def ask_anysecret(question):
    prompt = f"### Instruction:\n{question}\n\n### Response:\n"
    inputs = tokenizer(prompt, return_tensors="pt").to(device)
    
    with torch.no_grad():
        outputs = model.generate(**inputs, max_new_tokens=256, temperature=0.1)
    
    response = tokenizer.decode(outputs[0], skip_special_tokens=True)
    return response.split("### Response:\n")[-1].strip()

# Example usage
print(ask_anysecret("How do I set a secret using anysecret CLI?"))

Environmental Impact

• Hardware Type: NVIDIA RTX 3090 / A6000
• Hours used: ~2-4 hours per training run
• Training Framework: PyTorch with PEFT and BitsAndBytes
• Quantization: 4-bit NF4 for memory efficiency

Technical Specifications

Model Architecture and Objective

• Architecture: Llama-3.2 with LoRA adapters
• Objective: Causal language modeling for instruction following
• LoRA Configuration: Rank 16, Alpha 32, targeting attention and MLP layers
• Quantization: 4-bit NF4 with double quantization

Compute Infrastructure

Hardware

• NVIDIA RTX 3090 (24GB VRAM) for 3B models
• NVIDIA A6000 (48GB VRAM) for 13B models

Software

• PyTorch 2.0+
• Transformers 4.35+
• PEFT 0.6+
• BitsAndBytes 0.41+

Framework versions

• PEFT 0.17.1
• Transformers 4.35.0
• PyTorch 2.0.0
• BitsAndBytes 0.41.0