🩺 LogDoctor LogDoctor is a fine-tuned Llama 3.1 8B Instruct model specialized for software log analysis, debugging, error diagnosis, and root-cause investigation.

It is designed to turn messy technical logs and error traces into structured, actionable debugging assistance. ⚡ Quick Start Run LogDoctor locally with Ollama: ollama run anbukumaran/logdoctor Then provide a log, stack trace, or error message.

Example: Analyze this error and identify the likely root cause:

RuntimeError: CUDA out of memory. Tried to allocate 2.00 GiB… LogDoctor is designed to respond with practical reasoning around: - What went wrong - Likely root cause - Relevant error signals - Possible contributing factors - Recommended debugging steps - Potential fixes 🧠 What is LogDoctor? LogDoctor is a domain-specialized language model created by fine-tuning:

Base Model: meta-llama/Meta-Llama-3.1-8B-Instruct

The model was trained using a LoRA/QLoRA fine-tuning pipeline with Unsloth on a synthetic dataset designed around software-log and debugging scenarios.

The goal was not to create a general-purpose chatbot, but to specialize an existing instruction-following model toward a focused developer workflow: Raw Logs / Errors ↓ LogDoctor ↓ Error Understanding ↓ Root-Cause Analysis ↓ Actionable Debugging Guidance 🎯 Intended Use LogDoctor is intended for: - Application log analysis - Error diagnosis - Stack-trace interpretation - Debugging assistance - Root-cause investigation - Developer troubleshooting - Understanding unfamiliar error messages - Turning technical logs into structured explanations 🧪 Training The controlled fine-tuning experiment used: Component Value Base model meta-llama/Meta-Llama-3.1-8B-Instruct Fine-tuning method LoRA / QLoRA Training framework Unsloth Training examples 280 Validation examples 100 Held-out test examples 120 Epochs 3 Optimizer updates 210

The dataset was synthetically generated for the LogDoctor task and split into training, validation, and held-out test partitions. 💻 Running Locally Ollama allows LogDoctor to be used locally without requiring a hosted inference API. ollama run anbukumaran/logdoctor For programmatic usage, the Ollama API can also be used. Python from ollama import chat

response = chat( model=“anbukumaran/logdoctor”, messages=[ { “role”: “user”, “content”: “”“Analyze this error:ConnectionError: Connection refused on port 5432.Identify the likely root cause and give me a debugging plan.”“” } ], )

print(response.message.content) 🔬 Example Workflow Input ERROR: Database connection failed psycopg2.OperationalError: could not connect to server: Connection refused Is the server running on host “localhost”? Ask Analyze this log.

Give me: 1. The likely root cause 2. Evidence from the log 3. What I should check first 4. Possible fixes LogDoctor is designed to transform this type of raw diagnostic information into a structured troubleshooting response. ⚠️ Limitations LogDoctor is a specialized fine-tuned language model, not a replacement for actual observability, monitoring, debugging tools, or human engineering judgment.

It may produce incorrect diagnoses, especially when: - Logs are incomplete - Important context is missing - Multiple failures are present - The underlying issue is environment-specific - The error requires information unavailable in the supplied logs Always verify suggested fixes against the actual system. 🏗️ Model Lineage Meta Llama 3.1 8B Instruct │ ▼ LoRA / QLoRA │ ▼ Synthetic Debugging Dataset │ ▼ LogDoctor │ ┌─────┴─────┐ ▼ ▼ Hugging Face Ollama 📦 Model Hugging Face:

Anbukumaran1/LogDoctor-Llama-3.1-8B

Ollama:

anbukumaran/logdoctor 👨‍💻 Author Built and fine-tuned by Anbukumaran.

LogDoctor is part of an ongoing exploration of practical, locally deployable, domain-specialized language models.