From Signal Degradation to Computation Collapse: Uncovering the Two Failure Modes of LLM Quantization
This provides a diagnostic framework for improving quantization efficiency in LLM deployment, though it is incremental as it builds on existing PTQ methods.
The paper tackles the problem of catastrophic performance drops in 2-bit quantization of Large Language Models by uncovering two distinct failure modes: Signal Degradation and Computation Collapse, and shows that targeted repair can mitigate the former but not the latter.
Post-Training Quantization (PTQ) is critical for the efficient deployment of Large Language Models (LLMs). While 4-bit quantization is widely regarded as an optimal trade-off, reducing the precision to 2-bit usually triggers a catastrophic ``performance cliff.'' It remains unclear whether the underlying mechanisms differ fundamentally. Consequently, we conduct a systematic mechanistic analysis, revealing two qualitatively distinct failure modes: Signal Degradation, where the computational patterns remain intact but information precision is impaired by cumulative error; and Computation Collapse, where key components fail to function, preventing correct information processing and destroying the signal in the early layers. Guided by this diagnosis, we conduct mechanism-aware interventions, demonstrating that targeted, training-free repair can mitigate Signal Degradation, but remains ineffective for Computation Collapse. Our findings provide a systematic diagnostic framework for PTQ failures and suggest that addressing Computation Collapse requires structural reconstruction rather than mere compensation.