For example, we achieve a Wikitext2 perplexity of 5.95 on the LLaMA2-70B

Backward propagation of errors (backpropagation) is a method to minimize objective functions (e.g., loss functions) of deep neural networks by identifying optimal

Long-context Large Language Models (LLMs) face significant memory bottlenecks during inference due to the linear growth of key-value (KV) cache with sequence length. While individual optimization techniques like KV cache quantization, chunked prefill, and model weight quantization have shown promise, their joint effects and optimal configurations for edge deployment remain underexplored. We introduce KV Pareto, a systems-level framework that systematically maps the trade-off frontier between total memory consumption and task accuracy across these three complementary optimization techniques. Our framework evaluates multiple LLM architectures (Qwen, Llama, Mistral) with varying KV quantization schemes (int2/4/8, mixed-precision), granularities (per-token, per-tensor, per-block), and 4-bit weight quantization via AWQ. Our framework identifies model-specific Pareto-optimal configurations that achieve 68-78% total memory reduction with minimal (1-3%) accuracy degradation on long-context tasks. We additionally verify the selected frontiers on additional benchmarks of Needle-in-a-Haystack, GSM8k and MMLU as well as extended context lengths of up to 128k to demonstrate the practical need of joint optimization for efficient LLM inference.

Quantization of neural networks provides benefits of inference in less compute and memory requirements. Previous work in quantization lack two important aspects which this work provides. First almost all previous work in quantization used a non-differentiable approach and for learning; the derivative is usually set manually in backpropogation which make the learning ability of algorithm questionable, our approach is not just differentiable, we also provide proof of convergence of our approach to the optimal neural network. Second previous work in shift/logrithmic quantization either have avoided activation quantization along with weight quantization or achieved less accuracy. Learning logrithmic quantize values of form $2^n$ requires the quantization function can scale to more than 1 bit quantization which is another benifit of our quantization that it provides $n$ bits quantization as well. Our approach when tested with image classification task using imagenet dataset, resnet18 and weight quantization only achieves less than 1 percent accuracy compared to full precision accuracy while taking only 15 epochs to train using shift bit quantization and achieves comparable to SOTA approaches accuracy in both weight and activation quantization using shift bit quantization in 15 training epochs with slightly higher(only higher cpu instructions) inference cost compared to 1 bit quantization(without logrithmic quantization) and not requiring any higher precision multiplication.

The rapid advancement of large language models (LLMs) has exacerbated the memory bottleneck due to the widening gap between model parameter scaling and hardware capabilities. While post-training quantization techniques effectively reduce memory overhead, existing methods predominantly rely on static quantization strategies, which struggle to adapt to dynamic workloads. To address this, we propose FlexQuant, a dynamic precision-switching framework that optimizes the trade-off between inference speed and accuracy. Leveraging model perplexity entropy and Kullback-Leibler divergence, FlexQuant enables fine-grained, layer-wise mixed-precision quantization and dynamically adjusts bit-widths during each token generation. FlexQuant provides a comprehensive analysis of quantization strategies, introduces a precision requirement model for optimal switching, and implements efficient fine-grained precision management. Evaluations demonstrate that FlexQuant achieves a 1.3x end-to-end speedup across diverse language tasks with negligible accuracy loss introduced. This framework offers a flexible and adaptive solution for efficient LLM deployment. Code is released at https://github.com/ZongwuWang/FlexQuant.git.

For example, we achieve a Wikitext2 perplexity of 5.95 on the LLaMA2-70B

Diffusion-based large language models (DLLMs) have shown promise for non-autoregressive text generation, but their deployment is constrained by large model sizes and heavy computational costs. Post-training quantization (PTQ), a widely used method for compressing and accelerating Large Language Models (LLMs), suffers from severe accuracy degradation and reduced generalization performance when directly applied to DLLMs (e.g., AWQ suffers a 16% accuracy drop on LLADA under W4A4). This paper explores how DLLMs' key mechanisms - dynamic masking, iterative generation, bidirectional attention - clash with quantization. We identify three core issues: 1) Iterative generation and dynamic masking ratios lead to distinct token distributions across decoding steps, which are not adequately captured by existing PTQ calibration methods; 2) Quantization errors are accumulated and amplified progressively during iteration in DLLMs, causing quantized models to perform worse as decoding steps progress; 3) Unmasked tokens stabilize while masked remain probabilistic, making overall feature distribution incompatible with existing PTQ methods. To address these issues, we propose DLLMQuant, a PTQ framework tailored for DLLMs, which incorporates three novel techniques: 1) Temporal-Mask Adaptive Sampling (TMAS), a calibration method that accounts for both time and mask factors, with the capacity to capture distributions across timesteps. 2) Interaction-Aware Activation Quantization (IA-AQ), which utilizes bidirectional attention's interaction signals to dynamically allocate quantization resources. 3) Certainty-Guided Quantization (CGQ), which integrates mask status and token scores as key weighting criteria into error compensation, making weight quantization more suitable for DLLMs. Experiments show that DLLMQuant achieves significant performance gains while enhancing efficiency.

The accuracy drop varies a lot under different benchmarks and scenarios.

Recent advances in Automatic Speech Recognition (ASR) have demonstrated remarkable accuracy and robustness in diverse audio applications, such as live transcription and voice command processing. However, deploying these models on resource-constrained edge devices (e.g., IoT device, wearables) still presents substantial challenges due to strict limits on memory, compute and power. Quantization, particularly Post-Training Quantization (PTQ), offers an effective way to reduce model size and inference cost without retraining. Despite its importance, the performance implications of various advanced quantization methods and bit-width configurations on ASR models remain unclear. In this work, we present a comprehensive benchmark of eight state-of-the-art (SOTA) PTQ methods applied to two leading edge-ASR model families, Whisper and Moonshine. We systematically evaluate model performances (i.e., accuracy, memory I/O and bit operations) across seven diverse datasets from the open ASR leader-board, analyzing the impact of quantization and various configurations on both weights and activations. Built on an extension of the LLM compression toolkit, our framework integrates edge-ASR models, diverse advanced quantization algorithms, a unified calibration and evaluation data pipeline, with detailed analysis tools. Our results characterize the trade-offs between efficiency and accuracy, demonstrating that even $3$-bit quantization can succeed on high capacity models when using advanced PTQ techniques. These findings provide valuable insights for optimizing ASR models on low-power, always-on edge devices.