Beyond Scalar Sensitivity: Activation-Aware Mixed-Precision LLM Quantization with Cross-Layer Refinement

A self-hosted assistant is too expensive because its model uses most of the GPU memory.

What you would change

Try the paper's supported lower-precision weights on a copy of your serving setup.

Try the paper's supported lower-precision weights on a copy of your serving setup.

Compare against

The same model at its original precision, on the same GPU.

1. Locate the paper's implementation and check its license, model support, data needs, and hardware requirements. Stop here if you cannot obtain the required artifacts.
2. Save a small set of representative product tasks and expected outcomes. Keep evaluation tasks out of any training or tuning, and choose quality and cost limits before running.
3. Run the baseline and the proposed change on identical inputs. Keep other settings fixed and repeat paired runs to expose variation.
4. Save per-task outputs, failures, timings, and all usage costs. Review regressions before expanding the test.

What to measure

Peak memory, task pass rate, latency, and total serving cost at the same context and load.

When to keep it

Keep the change only if it meets your preselected quality and latency limits and improves the metric you care about after setup and operating costs. Otherwise retain the baseline. This is an exploration plan, not a validated recommendation.

recommendation

test

research area

quantization

technical change

Activation-aware mixed-precision quantization (CASA)

published

2026-09-22

The perplexity advantage is concentrated at ultra-low bit-widths, so a constrained-infrastructure engineer should validate it on their own target model and bit budget before displacing simpler, well-supported quantizers.

1. claim 01Paperraft inference
   Read the brief interpretation

   CASA replaces scalar Hessian sensitivity proxies in MCKP-based mixed-precision bit allocation with a Kronecker-factored activation-aware metric solved in closed form, followed by a cross-layer local search scored on end-to-end loss. The cost is additional quantization-time computation: calibration data, Hessian factor estimation per module, and an iterative local search that evaluates candidate bit assignments through full model forward passes, plus engineering effort beyond off-the-shelf GPTQ/AWQ pipelines. It can fail when calibration activations are unrepresentative, when the local search overfits the calibration loss, or when the model's modules are well-conditioned, in which case the method provably reduces to the cheaper scalar proxy it replaces.

   This is a Paperraft inference. It is not pinned to a verified PDF passage.

   Inspect the source
2. claim 02Paperraft inference

   CASA achieves lower perplexity than scalar-proxy mixed-precision baselines across bit budgets, with the largest gains below 3 bits per weight; zero-shot accuracy gains correlate with the model's average Hessian condition number. No multiplicative improvement factor is reported.

   This is a Paperraft inference. It is not pinned to a verified PDF passage.

   Inspect the source