Breaking the 1.58-bit Barrier for Ternary LLMs [LAB]
Explores extreme quantization techniques relevant to low-bitwidth local inference and hardware efficiency. https://arxiv.org/abs/2609.16338 (arxiv.org)
Training a 4B model to produce 81% faster query plans than Postgres [LAB]
Demonstrates specialized small-model training for specific reasoning tasks, aligning with interests in model architecture and performance. https://rohanbansal.com/qorl (rohanbansal.com)