{"id":"b0dac79a-47fb-45a3-8443-36907ceb095d","arxiv_id":"2606.00957","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"A boundary-protection PTQ strategy for Wan2.1-T2V-14B matches BF16 VBench performance by retaining boundary blocks in higher precision and quantizing the rest to W8A8 HiF8.","lead":"This paper presents a post-training quantization method for a 14-billion-parameter text-to-video diffusion model that keeps the first two and last three transformer blocks in BF16 precision while quantizing the middle blocks to W8A8 HiFloat8. A smart generalist might read it to learn practical techniques for running large generative AI models more efficiently on specialized NPUs with minimal quality impact.","discovery_kind":"new_method","skeptic_critique":{"model":"grok-4.3","headline":"Per-block activation statistics may vary with prompts, so fixed protection of blocks 1-2 and 38-40 may not generalize.","rationale":"The reader's weakest assumption correctly isolates the stability of the block-selection rule. Because the performance numbers are reported only on the calibration prompts, confirming that the same blocks remain extreme on held-out prompts is the minimal check needed to support the claim beyond the 5-prompt set.","tokens_in":1724,"tokens_out":313,"duration_ms":16218,"concrete_test":"Recompute the per-block activation statistics (range, 99th-percentile magnitude, or kurtosis) on a disjoint set of 20 VBench prompts; verify whether the same five blocks remain the clear outliers. If the outlier set shifts by more than one block, the fixed boundary-protection rule is prompt-dependent.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central claim rests on a data-driven choice to keep exactly the first two and last three of the 40 WanAttentionBlocks in BF16 while quantizing the middle 35 to W8A8 HiF8. This choice is justified by a per-block analysis of activation statistics, yet the paper reports results only on the same 5-prompt set used for that analysis. Diffusion transformer activations are known to shift with prompt content, text embeddings, and timestep. If the blocks exhibiting the most extreme ranges or outliers change under different prompts, the protection mask becomes suboptimal and the observed parity with BF16 may not hold.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.3","summary":"The paper introduces a post-training quantization (PTQ) technique for the 14B-parameter Wan2.1-T2V text-to-video diffusion transformer targeting W8A8 HiFloat8 on Ascend NPUs. Through per-block activation analysis on the 40 WanAttentionBlocks, it identifies that boundary blocks have distinct statistics and proposes protecting the first two and last three blocks in BF16 while quantizing the middle 35 blocks. The method is shown to match or exceed BF16 performance on all five VBench dimensions using five prompts, supported by an ablation study on protection configurations and analysis of QAT.","tokens_in":1872,"tokens_out":507,"duration_ms":23933,"significance":"Should the boundary-protection strategy prove robust across diverse prompts and models, this work would be significant for efficient deployment of large-scale video generation models on specialized hardware. It provides an empirical solution to the heterogeneous activation issue in DiTs without full model retraining. The systematic analysis and ablation offer practical insights, though the small evaluation scale limits broader claims.","major_comments":[{"comment":"Abstract and Evaluation: The central claim of matching BF16 performance 'indicating no measurable accuracy loss' rests on results from only five prompts (Abstract). Since the per-block activation analysis used to select exactly the first two and last three blocks for protection was performed on the same prompt set, the evaluation does not test generalization; prompt-dependent shifts in activation ranges could render the fixed protection mask suboptimal.","section":"Abstract and Evaluation"},{"comment":"Per-block analysis section: The manuscript does not clarify if the activation statistics for block selection were computed on a held-out calibration set distinct from the five evaluation prompts. This detail is load-bearing for assessing whether the observed parity is due to the method or to the data-driven choice being tuned to the test prompts.","section":"Per-block analysis section"}],"minor_comments":[{"comment":"The term 'HiFloat8 (HiF8)' should be defined more explicitly with its format details upon first use to aid readers unfamiliar with the Ascend-specific format.","section":"Notation"},{"comment":"The per-block activation range plots (presumably in the analysis section) would benefit from error bars across multiple runs or prompts to show variability.","section":"Figures"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":{"model":"grok-4.3","summary":"We thank the referee for the constructive comments highlighting the evaluation scale and calibration details. We address each point below and will revise the manuscript for greater clarity on these aspects.","responses":[{"response":"We agree the evaluation uses only five prompts and that block selection derives from analysis on this same set. The abstract already qualifies results as holding 'within the 5-prompt evaluation set.' This reflects standard PTQ practice where calibration data matches the target distribution. The ablation over protection configurations shows the selected boundaries yield the highest score on these prompts. We will revise to explicitly discuss the limited scope and lack of generalization testing, without claiming broader robustness.","revision_made":"partial","referee_comment":"[Abstract and Evaluation] Abstract and Evaluation: The central claim of matching BF16 performance 'indicating no measurable accuracy loss' rests on results from only five prompts (Abstract). Since the per-block activation analysis used to select exactly the first two and last three blocks for protection was performed on the same prompt set, the evaluation does not test generalization; prompt-dependent shifts in activation ranges could render the fixed protection mask suboptimal."},{"response":"The activation statistics were computed on the same five evaluation prompts; the manuscript does not describe a distinct held-out calibration set. We will revise the per-block analysis section to state this explicitly. The data-driven selection is thus tuned to the evaluated prompts, which is a limitation. However, the systematic per-block analysis and ablation still provide evidence that boundary blocks exhibit distinct statistics warranting protection. The revision will clarify this without overstating generalization.","revision_made":"yes","referee_comment":"[Per-block analysis section] Per-block analysis section: The manuscript does not clarify if the activation statistics for block selection were computed on a held-out calibration set distinct from the five evaluation prompts. This detail is load-bearing for assessing whether the observed parity is due to the method or to the data-driven choice being tuned to the test prompts."}],"tokens_in":1426,"tokens_out":425,"duration_ms":23521,"standing_objections":[]},"desk_editor":{"model":"grok-4.3","letter":"The core result is that a per-block activation scan on the 40 WanAttentionBlocks revealed clear differences at the start and end, so they protect blocks 1-2 and 38-40 in full precision and quantize the middle 35. That choice, backed by an ablation of four protection setups, produces VBench scores that match or slightly beat BF16 on the five prompts they tried.\n\nWhat stands out is the systematic activation analysis that led to the boundary rule instead of a blanket approach. They also checked QAT as a follow-up and noted when it does not beat plain PTQ on single-card hardware. That kind of targeted engineering detail is useful for people who actually ship these models.\n\nThe soft spot is the evaluation size. Five prompts with no error bars or statistical checks leaves open the question of whether the same blocks stay the outliers when the prompt, timestep, or text embedding changes. The stress-test note is on point here: if activation ranges shift, the fixed protection mask could lose its edge. The work is also tied to one model and one NPU platform, so the numbers are not meant to generalize broadly.\n\nThis is for engineers who need to run large video diffusion models under tight memory on Ascend hardware and want a concrete recipe plus the analysis that produced it. A reader already working on DiT quantization or deployment will pick up the per-block method and the ablation results.\n\nIt is worth sending to peer review. The empirical claim is narrow but cleanly supported by the ablation, and the activation analysis gives other groups something to test on their own models.","headline":"This paper shows a data-driven way to quantize most blocks of a 14B video DiT to W8A8 HiF8 while keeping boundary blocks in BF16, and it matches the baseline on their five-prompt test.","tokens_in":2343,"tokens_out":416,"would_cite":false,"duration_ms":12910,"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":"Protecting boundary blocks in full precision while quantizing the middle ones enables W8A8 HiFloat8 inference on a 14B text-to-video diffusion transformer with no measurable accuracy loss.","keywords":["post-training quantization","text-to-video diffusion","boundary protection","HiFloat8","diffusion transformers","Wan2.1-T2V","activation analysis"],"falsifier":"Running the same W8A8 model on a larger prompt set or different video tasks and measuring a clear drop below the BF16 VBench scores would show the protection choice does not generalize.","tokens_in":2616,"feed_emoji":"","tokens_out":674,"duration_ms":14618,"temperature":0.7,"pith_summary":"The paper shows that activation statistics vary sharply across the 40 blocks of a large video diffusion transformer, with the first and last blocks differing from the middle ones. By keeping the first two and last three blocks in BF16 and applying W8A8 HiFloat8 only to the central 35 blocks, the method matches or slightly exceeds the full-precision baseline on all five VBench dimensions. An ablation across protection schemes confirms that the full boundary set gives the best average score. The work also notes conditions where adding quantization-aware training does not improve results on single-card hardware.","feed_headline":"Boundary protection enables lossless 8-bit quantization of 14B video model","feed_subtitle":"Keeping first two and last three blocks in BF16 while quantizing the rest matches full-precision VBench scores.","key_machinery":"Boundary-protection strategy that uses per-block activation analysis to select which transformer blocks remain in BF16.","core_discovery":"A boundary-protection post-training quantization strategy retains the first two and last three WanAttentionBlocks in BF16 while quantizing the remaining 35 blocks to W8A8 HiFloat8; this configuration matches or marginally exceeds the BF16 baseline across all VBench dimensions on the 14B Wan2.1-T2V-14B model within the 5-prompt test set, and an ablation verifies that this exact protection choice outperforms the other three configurations examined.","pith_inferences":["The heterogeneous block statistics may appear in other diffusion transformer families, suggesting the protection pattern could transfer after similar analysis.","Extending the evaluation to hundreds of prompts would test whether the no-loss result holds outside the current small set.","The approach could be combined with other compression techniques such as pruning or distillation for further efficiency gains."],"forward_implications":["Quantizing 35 of 40 blocks reduces memory footprint and inference cost on NPUs while preserving output quality.","The same per-block analysis can be repeated on other large DiT models to identify their own boundary blocks.","QAT provides no consistent gain over this PTQ recipe on single-card hardware.","Full boundary protection is the configuration that maximizes average VBench score among the four tested options."],"fun_headline_variants":["Boundary blocks kept in BF16 during 14B video DiT W8A8 HiF8 quantization","Per-block stats motivate protecting first two and last three blocks in BF16","35 middle blocks quantized to W8A8 HiF8 with boundary blocks in BF16","Activation analysis across 40 blocks guides boundary protection for PTQ"],"cache_read_input_tokens":2112,"weakest_assumption_plain":"The activation statistics measured on the 40 blocks remain stable enough that protecting exactly the first two and last three blocks will continue to work for other prompts and model configurations.","fun_headline_variants_meta":{"raw":{"variants":["Boundary blocks kept in BF16 during 14B video DiT W8A8 HiF8 quantization","Per-block stats motivate protecting first two and last three blocks in BF16","35 middle blocks quantized to W8A8 HiF8 with boundary blocks in BF16","Activation analysis across 40 blocks guides boundary protection for PTQ"]},"model":"grok-4.3","cost_usd":0.004395,"raw_usage":{"total_tokens":2216,"prompt_tokens":701,"num_sources_used":0,"completion_tokens":86,"cost_in_usd_ticks":43949500,"prompt_tokens_details":{"text_tokens":701,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":1429,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":701,"tokens_out":86,"duration_ms":9811,"temperature":1.0,"reasoning_tokens":1429,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-06-28T17:46:51.685339+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"Running the same W8A8 model on a larger prompt set or different video tasks and measuring a clear drop below the BF16 VBench scores would show the protection choice does not generalize.","supporting_citations":[],"review_version":1}