REVIEW 2 cited by
Mamba-PTQ: Outlier Channels in Recurrent Large Language Models
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
read the original abstract
Modern recurrent layers are emerging as a promising path toward edge deployment of foundation models, especially in the context of large language models (LLMs). Compressing the whole input sequence in a finite-dimensional representation enables recurrent layers to model long-range dependencies while maintaining a constant inference cost for each token and a fixed memory requirement. However, the practical deployment of LLMs in resource-limited environments often requires further model compression, such as quantization and pruning. While these techniques are well-established for attention-based models, their effects on recurrent layers remain underexplored. In this preliminary work, we focus on post-training quantization for recurrent LLMs and show that Mamba models exhibit the same pattern of outlier channels observed in attention-based LLMs. We show that the reason for the difficulty of quantizing SSMs is caused by activation outliers, similar to those observed in transformer-based LLMs. We report baseline results for post-training quantization of Mamba that do not take into account the activation outliers and suggest first steps for outlier-aware quantization.
Forward citations
Cited by 2 Pith papers
-
Neural Weight Compression for Language Models
A single learned neural codec, trained once on real LLM weights, compresses Llama-scale models to 4-6 bits per weight with near-FP16 accuracy, beating hand-crafted quantization at those bitrates.
-
Quantizing Small-Scale State-Space Models for Edge AI
Quantization-aware training with a frozen state matrix lifts sequential MNIST accuracy from 40% under post-training quantization to 96%, and a heterogeneous precision scheme cuts memory by 6 times.
Discussion (0). Sign in to comment.