Uploaded March 2026 | Updated November 2026, 2 days ago
Stop overpaying for massive LLMsโitโs the fastest way to kill your inference performance and blow your infrastructure budget โ๏ธ
In this technical deep-dive, Legare Kerrison (Red Hat) reveals the engineering framework for quantization and building high-performance, task-specific agentic pipelines ๐ซ
Whatโs inside for Senior Engineers:
โ Model Compression (FP16 to INT4): The math of quantizationโhow to retain 99% accuracy while doubling your inference speed.
โ Production Inference with vLLM: Architecting low-latency stacks using open-source engines and SafeTensors.
โ Data Sovereignty & Local SLMs: Moving beyond cloud endpointsโrunning 8B models on-prem for total privacy and infrastructure control.
โ The Agentic Paradigm: Why chaining specialized "Small" models outperforms single giant LLMs in complex multi-step workflows.
Watch the full talk on our YouTube channel: JavaScript Conferences by GitNation ๐
Link in Bio ๐
#GitNation #AICodingSummit2026 #LLM #SLM #Quantization
Stop overpaying for massive LLMsโitโs the fastest way to kill your inference performance and blow your infrastructure budget โ๏ธ
In this technical deep-dive, Legare Kerrison (Red Hat) reveals the engineering framework for quantization and building high-performance, task-specific agentic pipelines ๐ซ
Whatโs inside for Senior Engineers:
โ Model Compression (FP16 to INT4): The math of quantizationโhow to retain 99% accuracy while doubling your inference speed.
โ Production Inference with vLLM: Architecting low-latency stacks using open-source engines and SafeTensors.
โ Data Sovereignty & Local SLMs: Moving beyond cloud endpointsโrunning 8B models on-prem for total privacy and infrastructure control.
โ The Agentic Paradigm: Why chaining specialized "Small" models outperforms single giant LLMs in complex multi-step workflows.
Watch the full talk on our YouTube channel: JavaScript Conferences by GitNation ๐
Link in Bio ๐
#GitNation #AICodingSummit2026 #LLM #SLM #Quantization










