{"id":"2ab3550f-7ce9-4443-a86b-cedfc09ef959","arxiv_id":"2606.29337","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":3.0,"correctness_risk":"high","formal_verification":"none","parameter_count":2,"one_line_summary":"W4A4 quantization via MixQ-style mixed precision, SmoothQuant-style smoothing, and dual-branch GEMM keeps Wan2.2-I2V-A14B within 2-3.5% of FP16 on VBench I2V metrics while beating a native HiFloat4 baseline.","lead":"This paper describes adapting LLM quantization techniques like mixed-precision outlier handling and per-channel smoothing to enable 4-bit weight and activation inference on a large image-to-video model. A smart generalist might read it to understand practical steps for reducing compute and memory costs in deploying video generation systems.","discovery_kind":"new_application","skeptic_critique":{"model":"grok-4.3","headline":"Calibration on OpenS2V-5M may not transfer to Wan2.2-I2V-A14B activation statistics","rationale":"The reader's weakest_assumption directly identifies the same transfer risk; the abstract (and likely the short challenge submission) supplies the performance numbers but no evidence that the calibration step generalizes to the target model.","tokens_in":1734,"tokens_out":312,"duration_ms":19622,"concrete_test":"Re-run the full pipeline with calibration batches drawn from Wan2.2-I2V-A14B inference on a held-out prompt set instead of OpenS2V-5M; recompute the official VBench I2V scores. If any axis falls outside the claimed 2-3.5% band relative to FP16, the transfer assumption fails.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim (VBench scores within 2-3.5% of FP16, outperforming native HiFloat4) rests on identifying heavy-tailed channels via OpenS2V-5M calibration, then applying per-channel smoothing plus dual-branch GEMM without retraining. For this to hold, the outlier channels and dynamic-range rebalancing must be representative of Wan2.2-I2V-A14B's actual inference activations. The described pipeline provides no cross-model validation or ablation showing that the identified channels match those arising during I2V generation, so any distribution shift would directly undermine the reported closeness to FP16.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.3","summary":"The manuscript describes a W4A4 quantization approach for the Wan2.2-I2V-A14B image-to-video model submitted to the ICME 2026 challenge. It adapts MixQ-style mixed precision for outliers and SmoothQuant-style per-channel smoothing, with calibration on OpenS2V-5M batches to identify heavy-tailed channels, followed by smoothing and dual-branch GEMM for inference. The key empirical claim is that on VBench I2V metrics, the quantized model stays within 2-3.5% of FP16 performance on most axes, improves motion smoothness, and outperforms a native HiFloat4 baseline which degrades by ~5%.","tokens_in":1889,"tokens_out":498,"duration_ms":22146,"significance":"If the reported performance holds under scrutiny, the work would provide a practical demonstration of effective 4-bit weight and activation quantization for a large-scale video generation model without retraining, which could have significant implications for deploying such models efficiently. The adaptation of LLM quantization techniques to this domain is a reasonable extension, but the absence of detailed methodology, ablations, or statistical validation limits the immediate impact.","major_comments":[{"comment":"Abstract: The central performance claims (within 2-3.5% of FP16 on VBench metrics) are presented without any implementation details, error bars, statistical tests, ablation results, or description of the exact calibration procedure, making it impossible to evaluate or reproduce the results.","section":"Abstract"},{"comment":"Abstract: The pipeline relies on the assumption that heavy-tailed activation channels identified via calibration on OpenS2V-5M batches are representative of those in Wan2.2-I2V-A14B during I2V generation; no cross-model validation or ablation is provided to support transfer of these statistics, which directly bears on whether the reported closeness to FP16 can be achieved.","section":"Abstract"}],"minor_comments":[{"comment":"Abstract: The term 'native HiFloat4 baseline' is not defined; clarify what this baseline entails and how it differs from the proposed pipeline.","section":"Abstract"}],"recommendation":"major_revision","confidential_remarks":"This appears to be a short challenge submission paper; the lack of technical depth may indicate it is intended as an extended abstract rather than a full manuscript."},"author_rebuttal":{"model":"grok-4.3","summary":"We thank the referee for the constructive feedback on our ICME 2026 challenge submission. This work is presented as a concise summary due to strict page limits, but we address the concerns about missing details and assumptions below. We clarify aspects of the methodology while noting constraints of the format.","responses":[{"response":"The submission format for the challenge requires a compact abstract-style summary, which limits inclusion of error bars, statistical tests, and full ablations. The manuscript does outline the core pipeline: MixQ-style mixed precision for outliers, SmoothQuant-style per-channel smoothing, block-wise HiF4 packing, and dual-branch GEMM. The calibration procedure is described as processing OpenS2V-5M batches to detect heavy-tailed channels before applying smoothing and mixed-precision inference. We agree additional quantitative validation would improve reproducibility and can expand the description of the calibration steps in a revised version if space allows.","revision_made":"partial","referee_comment":"[Abstract] Abstract: The central performance claims (within 2-3.5% of FP16 on VBench metrics) are presented without any implementation details, error bars, statistical tests, ablation results, or description of the exact calibration procedure, making it impossible to evaluate or reproduce the results."},{"response":"OpenS2V-5M was chosen as calibration data because it provides diverse, large-scale video content aligned with I2V tasks. The empirical results on VBench I2V metrics for the target Wan2.2-I2V-A14B model (within 2-3.5% of FP16 and outperforming the HiFloat4 baseline) provide indirect support for the effectiveness of the transferred statistics. We acknowledge that no explicit ablation on calibration dataset choice or cross-validation is included, which is a limitation of the concise format. The assumption is standard in post-training quantization but could be discussed more explicitly if a revision is requested.","revision_made":"no","referee_comment":"[Abstract] Abstract: The pipeline relies on the assumption that heavy-tailed activation channels identified via calibration on OpenS2V-5M batches are representative of those in Wan2.2-I2V-A14B during I2V generation; no cross-model validation or ablation is provided to support transfer of these statistics, which directly bears on whether the reported closeness to FP16 can be achieved."}],"tokens_in":1439,"tokens_out":511,"duration_ms":23478,"standing_objections":[]},"desk_editor":{"model":"grok-4.3","letter":"The main takeaway is that the authors combine two established LLM quantization ideas—MixQ mixed precision for outliers and SmoothQuant per-channel smoothing—plus block-wise HiF4 packing, then calibrate on OpenS2V-5M batches to run Wan2.2-I2V-A14B at W4A4. They report VBench I2V scores within 2-3.5% of FP16 on most axes and better motion smoothness than a native HiFloat4 baseline.\n\nThe work does a clean job of targeting a real 14B video model and stating the high-level pipeline without claiming new primitives. The dual-branch GEMM for keeping outlier columns in higher precision is a practical choice for the feed-forward layers.\n\nThe soft spots are the missing substance. The text gives no description of the exact outlier threshold, how the smoothing scales are computed, what the block packing looks like in code, or any ablation showing that the heavy-tailed channels found on OpenS2V-5M match those arising during actual I2V generation. The stress-test point about possible distribution shift is not addressed, and there are no error bars, multiple seeds, or cross-model checks. Without those, the closeness to FP16 cannot be evaluated.\n\nThis paper is for engineers who need to deploy exactly this model under tight memory limits. A reader already working on similar video diffusion quantization might extract the high-level recipe, but the lack of detail limits reuse.\n\nI would send it to peer review so the authors can add the implementation steps and any validation that would let others judge whether the calibration transfers.","headline":"This paper applies MixQ and SmoothQuant to Wan2.2-I2V-A14B with block packing but supplies almost no implementation details or validation.","tokens_in":2371,"tokens_out":402,"would_cite":false,"duration_ms":28907,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.3","headline":"A W4A4 quantization pipeline for Wan2.2-I2V-A14B stays within 2-3.5 percent of FP16 quality on VBench metrics.","keywords":["W4A4 quantization","video generation model","activation smoothing","mixed precision","inference efficiency","VBench metrics","Wan2.2-I2V-A14B"],"falsifier":"Evaluating the quantized model on a new set of video prompts outside the calibration distribution and checking whether the quality degradation exceeds 3.5 percent on VBench metrics.","tokens_in":2648,"feed_emoji":"🎥","tokens_out":646,"duration_ms":37809,"temperature":0.7,"pith_summary":"This paper shows how to quantize a large image-to-video model to 4-bit weights and activations while keeping output quality close to the original 16-bit version. The method uses calibration to find channels with extreme activation values, then smooths their range and routes outliers through a higher-precision path in matrix multiplications. On standard video quality tests, most metrics drop only a small amount and motion smoothness even improves. A basic 4-bit approach without these steps loses about 5 percent across the board. The work demonstrates that existing ideas from language model quantization can transfer to video generation models for efficient inference.","feed_headline":"W4A4 method keeps video model quality within 3.5% of FP16","feed_subtitle":"Mixed precision for outliers and channel smoothing allow the pipeline to beat a standard 4-bit baseline on motion and other metrics.","key_machinery":"Dual-branch GEMM that keeps outlier columns in higher precision while applying W4A4 to the rest, after per-channel smoothing of heavy-tailed activations identified by calibration.","core_discovery":"Adapting mixed-precision handling of activation outliers via dual-branch GEMM and per-channel smoothing, combined with block-wise HiF4 packing, allows the 4-bit pipeline to match FP16 performance within 2-3.5 percent on most VBench I2V axes and to improve motion smoothness, while a standard HiFloat4 implementation falls about 5 percent behind FP16 on all scores.","pith_inferences":["This indicates that activation outlier management is critical for maintaining quality in quantized video diffusion models.","Similar calibration-based smoothing might reduce the quality gap in other generative tasks like text-to-video.","Deployment on edge devices could become feasible if the memory savings scale with model size."],"forward_implications":["The quantized model supports inference with lower memory and compute demands.","Motion smoothness can improve rather than degrade under this quantization scheme.","The approach requires only calibration data and no additional model training.","Outperforms a direct application of HiFloat4 without the mixed-precision and smoothing steps."],"fun_headline_variants":["W4A4 quantization on Wan2.2 holds within 3.5% of FP16","Channel smoothing aids W4A4 near FP16 on I2V metrics","Mixed precision for outliers keeps 4-bit video close to FP16","HiF4 packing with dual GEMM supports W4A4 on video model","4-bit pipeline stays within 2-3.5% of FP16 on VBench"],"cache_read_input_tokens":2112,"weakest_assumption_plain":"That calibration batches from OpenS2V-5M capture the activation statistics well enough for the target model without needing retraining or more data.","fun_headline_variants_meta":{"raw":{"variants":["W4A4 quantization on Wan2.2 holds within 3.5% of FP16","Channel smoothing aids W4A4 near FP16 on I2V metrics","Mixed precision for outliers keeps 4-bit video close to FP16","HiF4 packing with dual GEMM supports W4A4 on video model","4-bit pipeline stays within 2-3.5% of FP16 on VBench"]},"model":"grok-4.3","cost_usd":0.007959,"raw_usage":{"total_tokens":3632,"prompt_tokens":682,"num_sources_used":0,"completion_tokens":106,"cost_in_usd_ticks":79587000,"prompt_tokens_details":{"text_tokens":682,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":2844,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":682,"tokens_out":106,"duration_ms":29481,"temperature":1.0,"reasoning_tokens":2844,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-06-30T07:25:45.756546+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"Evaluating the quantized model on a new set of video prompts outside the calibration distribution and checking whether the quality degradation exceeds 3.5 percent on VBench metrics.","supporting_citations":[],"review_version":1}