Pere Martra is an experienced Applied AI Engineer, known for building efficient and scalable AI solutions. With over a decade of experience in the technology industry, Pere brings a practical and hands-on style to his teaching. He translates complex model optimisation techniques into clear, actionable steps that help engineers build better and more cost-effective AI systems.

• PART 1: FOUNDATIONS 1 WHY REARCHITECTING LLMS MATTERS 2 AN END-TO-END REARCHITECTING PROJECT 3 A BLUEPRINT TO MODERN TRANSFORMERS PART 2: HANDS-ON OPTIMIZATION 4 BUILDING SMALLER AND FASTER LLMS WITH DEPTH PRUNING 5 SHAPING GLU ARCHITECTURES VIA WIDTH PRUNING 6 RECOVERING KNOWLEDGE 7 SPECIALIZATION TUNING 8 ATTENTION OPTIMIZATION 9 DYNAMIC PRUNING FOR ADAPTIVE INFERENCE PART 3: BEYOND THE BLACK BOX: FAIRNESS, VISUALIZATION, AND APPLICATION 10 EXPLORING THE BLACK BOX 11 OPTIMIZING WHILE ELIMINATING BIASES 12 CAPSTONE PROJECT - REPLACING AGENTIC API CALLS WITH A SPECIALIZED SLM APPENDICES APPENDIX A: EFFICIENCY STRATEGIES IN SLMS APPENDIX B: CAPABILITIES EVALUATION WITH LM-EVALUATION-HARNESS APPENDIX C: ENERGY CONSUMPTION EVALUATION WITH CODE CARBON APPENDIX D: INFERENCE PERFORMANCE EVALUATION

Pere Martra is an experienced Applied AI Engineer, known for building efficient and scalable AI solutions. With over a decade of experience in the technology industry, Pere brings a practical and hands-on style to his teaching. He translates complex model optimisation techniques into clear, actionable steps that help engineers build better and more cost-effective AI systems.

• PART 1: FOUNDATIONS 1 WHY REARCHITECTING LLMS MATTERS 2 AN END-TO-END REARCHITECTING PROJECT 3 A BLUEPRINT TO MODERN TRANSFORMERS PART 2: HANDS-ON OPTIMIZATION 4 BUILDING SMALLER AND FASTER LLMS WITH DEPTH PRUNING 5 SHAPING GLU ARCHITECTURES VIA WIDTH PRUNING 6 RECOVERING KNOWLEDGE 7 SPECIALIZATION TUNING 8 ATTENTION OPTIMIZATION 9 DYNAMIC PRUNING FOR ADAPTIVE INFERENCE PART 3: BEYOND THE BLACK BOX: FAIRNESS, VISUALIZATION, AND APPLICATION 10 EXPLORING THE BLACK BOX 11 OPTIMIZING WHILE ELIMINATING BIASES 12 CAPSTONE PROJECT - REPLACING AGENTIC API CALLS WITH A SPECIALIZED SLM APPENDICES APPENDIX A: EFFICIENCY STRATEGIES IN SLMS APPENDIX B: CAPABILITIES EVALUATION WITH LM-EVALUATION-HARNESS APPENDIX C: ENERGY CONSUMPTION EVALUATION WITH CODE CARBON APPENDIX D: INFERENCE PERFORMANCE EVALUATION

Pere Martra is an experienced Applied AI Engineer, known for building efficient and scalable AI solutions. With over a decade of experience in the technology industry, Pere brings a practical and hands-on style to his teaching. He translates complex model optimisation techniques into clear, actionable steps that help engineers build better and more cost-effective AI systems.

• PART 1: FOUNDATIONS 1 WHY REARCHITECTING LLMS MATTERS 2 AN END-TO-END REARCHITECTING PROJECT 3 A BLUEPRINT TO MODERN TRANSFORMERS PART 2: HANDS-ON OPTIMIZATION 4 BUILDING SMALLER AND FASTER LLMS WITH DEPTH PRUNING 5 SHAPING GLU ARCHITECTURES VIA WIDTH PRUNING 6 RECOVERING KNOWLEDGE 7 SPECIALIZATION TUNING 8 ATTENTION OPTIMIZATION 9 DYNAMIC PRUNING FOR ADAPTIVE INFERENCE PART 3: BEYOND THE BLACK BOX: FAIRNESS, VISUALIZATION, AND APPLICATION 10 EXPLORING THE BLACK BOX 11 OPTIMIZING WHILE ELIMINATING BIASES 12 CAPSTONE PROJECT - REPLACING AGENTIC API CALLS WITH A SPECIALIZED SLM APPENDICES APPENDIX A: EFFICIENCY STRATEGIES IN SLMS APPENDIX B: CAPABILITIES EVALUATION WITH LM-EVALUATION-HARNESS APPENDIX C: ENERGY CONSUMPTION EVALUATION WITH CODE CARBON APPENDIX D: INFERENCE PERFORMANCE EVALUATION