{"id":"e790db35-f203-4b44-a0e5-85290b929e65","arxiv_id":"2611.03125","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":6,"one_line_summary":"A 3B student distilled with SurgPace beats six baselines on 15 of 16 Surg-396K metrics and roughly matches or exceeds its 13B teacher on short-answer and description tasks.","lead":"SurgPace trains a 3-billion-parameter surgical vision-language student by distilling a 13-billion-parameter teacher with a difficulty-aware curriculum and multi-level alignment. On the Surg-396K benchmark it outperforms six distillation baselines on most metrics while using about four times less memory and decoding about 1.4 times faster.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek/deepseek-v4-flash","headline":"The central empirical result is not yet trustworthy because the final configuration was selected on the same test split used to report the headline 0.6955; validation-based selection is the deciding check.","rationale":"The reader's verdict is CONDITIONAL, and I agree with that verdict. My most load-bearing concern is not the Eq. 10 min rule: even if s_i is a noisy difficulty signal, the full framework could still beat baselines through hierarchical alignment and progressive scheduling, and Tables 11-12 show that ordering direction matters. What actually threatens the reported numbers is that the final configuration was chosen from test-set performance. Section 4.4 highlights a 'final' row in Tables 11-16 based on best metrics on the fixed test split, with no validation-based selection and no seed averaging for the ablations. The headline 0.6955 is thus a selected maximum across roughly 20 single-seed configurations. This also weakens the curriculum evidence: Table 14's full-vs-no-curriculum gap (+0.0044 on ten metrics) is close to the seed standard deviation reported in Table 4 (0.0038 on sixteen metrics), so without multiple seeds the small curriculum contribution may be noise. I credit the paper for a reasonably controlled common protocol, a shared student architecture and initialization, three-seed robustness for the SurgPace-vs-EfficientVLM comparison, and the public code link; those mitigate but do not remove the selection problem. A validation-based reselection experiment would settle whether the central empirical claim survives, so I keep the reader's CONDITIONAL verdict unchanged.","tokens_in":29859,"tokens_out":12689,"duration_ms":82076,"concrete_test":"Take the already-trained variants behind Tables 11-16 (or retrain them), compute their ten-metric means on the 64,878 validation instances, choose the best configuration by validation only, then evaluate that single frozen configuration on the 64,856 test instances once. If the validation-selected model's 16-metric mean falls materially below 0.6955 or no longer exceeds EfficientVLM's corresponding number, the reported advantage is selection-driven; in any case, report how many configurations were evaluated on test and the omitted schedule constants (r_end, w_min, w_max, phase ratios, and beta endpoints).","verdict_should_be":"UNCHANGED","load_bearing_attack":"Tables 11-16 in Section 4.4 (complexity fusion, curriculum order, pacing, phase transition, and alignment) all report numbers on the fixed 64,856-example test split of Section 4.1, and each table marks the row that becomes the final configuration by best observed metrics under seed 42. No validation-set selection is reported; the 64,878 validation examples are never mentioned in the ablations. The headline 0.6955 and the 'best on 15 of 16 metrics' statement are therefore not a pre-specified comparison but the maximum of a small model-selection loop over roughly 20 single-seed configurations evaluated on the same test split used for the headline. This inflates the margin over baselines and makes the 'best student method' claim about a selected configuration rather than an independent estimate. It also affects the curriculum evidence: the full-vs-no-curriculum gain in Table 14 is +0.0044 on the ten ablated metrics, comparable to the 16-metric seed standard deviation of 0.0038 in Table 4, and it is itself selected on test. Limitations (i)-(iv) acknowledge teacher dependence, fixed scores, hardware scope, and audit design, but not test-based configuration selection.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek/deepseek-v4-flash","summary":"SurgPace proposes a curriculum-based distillation framework for compressing a 13B surgical multimodal LLM (EndoChat) into a 3B TinyLLaVA-style student built on Phi-2 and SigLIP. Sample complexity is scored offline from the teacher by taking the permissive minimum of two signals: the complement of the mean teacher top-1 token probability on the forced gold-answer sequence, and the Lorentz geodesic dispersion of teacher logits over that sequence. These fixed scores are converted into progress-dependent sample weights with an annealing schedule (easy-to-hard). The student objective combines cross-entropy with hierarchical alignment (temperature-scaled KL on a shared-vocabulary token support, centered logit L2, a learnable-head feature alignment, and entropy regularization), with a three-phase coefficient shifting emphasis from teacher imitation to task supervision. On a fixed split of the Surg-396K benchmark, the paper reports that the SurgPace student attains the highest unweighted average over 16 metrics (0.6955) among six distillation baselines and is best on 15 of 16 metrics, with a three-seed repeat against the strongest baseline (EfficientVLM: 0.6930±0.0038 vs 0.6221±0.0029), controlled efficiency measurements (~4x fewer parameters, 1.39x higher output throughput), an LLM-as-a-judge evaluation, a blinded clinical audit of detailed-description outputs, and ablations of the fusion rule, curriculum order, pacing function, phase transition, and alignment terms.","tokens_in":30238,"tokens_out":18450,"duration_ms":104111,"significance":"The paper's strengths are its controlled comparison design and its transparency: all student methods share the same architecture, teacher, initialization, split, and generation protocol; the strongest-baseline comparison is repeated over three seeds (Table 4); the difficulty-score sensitivity is quantified (Table 5); training budgets are reported (Table 17); the judge and audit prompts, sampling rules, and configuration hashes are given in the supplement; and the limitation list is honest about teacher dependence, fixed scores, hardware scope, and audit design. If the headline comparison holds, the result is a credible, practically relevant demonstration that a compact student can match or exceed a much larger teacher on several surgical dialogue paradigms at substantially lower cost. The principal weakness is evaluative rather than derivational: the final configuration is selected on the same test split that produces the headline number, so the exact magnitudes of the component contributions and the 0.6955 figure carry selection bias.","major_comments":[{"comment":"Configuration selection is performed on the same test split that produces the headline number. Every ablation table in §4.4 reports metrics on the fixed 64,856-example test split of §4.1, and the row adopted as the final configuration is in each case the best row under seed 42 on that split (s_min in Table 11, easy-to-hard in Table 12, linear in Table 13, full in Table 14, continuous in Table 15, full alignment in Table 16). The 64,878-example validation split is never used for selection. The headline 0.6955 and the 'best on 15 of 16 metrics' claim are therefore the maximum of a selection loop over 17 distinct single-seed configurations evaluated on that test split, not an independent estimate, and the expected optimism of such selection is of the same order as the reported across-seed variability (0.0038, Table 4), which is also the scale of several ablation differences (e.g., 0.0044, Table 14). Note that Table 4 repeats the already-selected configuration over seeds and therefore measures the stability of that configuration, not the bias of the selection procedure. The required corrective step is to select the configuration using the unused validation split and report the test numbers of the validation-selected configuration, or at minimum to show that the validation ranking of the ablation rows reproduces the adopted choices.","section":"§4.4 (Tables 11-16); §4.1-4.2","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The evidence that the complexity-aware curriculum itself improves distillation is within run-to-run noise. Removing the entire sample-level curriculum changes the ten-metric ablation mean from 0.6585 to 0.6541 (+0.0044, Table 14), while the only reported across-seed variability for an aggregate, the 16-metric mean, is ±0.0038 (Table 4). Several individual metric movements are below one seed standard deviation (e.g., Grounding QA AP50 0.9625 vs 0.9619 with a seed std of 0.0031; Single Phrase QA Acc 0.9004 vs 0.8928 with a seed std of 0.0101). Because the curriculum is the paper's stated primary contribution, the on/off comparison should be repeated over multiple seeds and reported with the same uncertainty convention as Table 4; with the current evidence, the curriculum's contribution and the selection noise are statistically indistinguishable.","section":"§4.4, Table 14; §4.3, Table 4","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The difficulty signal may partly measure teacher format agreement rather than content difficulty. Eq. (1) averages teacher top-1 probability on the gold token sequence under teacher-forcing, but the teacher's free-form answers differ in format from the gold for some paradigms (Table 2: EndoChat's Single Phrase QA accuracy is 0.5485 under the common protocol, which the paper attributes to answer-format sensitivity). For such samples, low teacher confidence on the forced gold tokens may reflect answer-style mismatch rather than content difficulty. The permissive min rule of Eq. (10) is a load-bearing modeling choice whose evidence is a small, test-selected gain (Table 11: s_min vs s_mean, 0.6585 vs 0.6567). Please provide a diagnostic that the scores track learnability rather than teacher style, for example the distribution of s_i per task family, or the correlation of s_i with per-sample student loss or error.","section":"§3.2, Eqs. (1)-(10); §4.3, Table 2","authors_acknowledge":null,"authors_acknowledge_p":null}],"minor_comments":[{"comment":"The 'best on 15 of 16 metrics' statement rests on single-seed runs for five of the six baselines; only EfficientVLM has repeated-seed numbers (Table 4). Please add seeds for at least the closest competing baseline or qualify the claim near the sentence that makes it.","section":"§4.3, Tables 2-3","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The limitations list (i)-(iv) does not acknowledge that the final configuration was selected on the test split, nor that the evaluation uses the authors' own teacher (EndoChat) and benchmark (Surg-396K); both facts should be stated so readers can calibrate the component-level claims.","section":"§5, Limitations","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The method name 'LLA V ADI' has broken spacing, and the 'LLaV A-...' naming is inconsistent throughout the text (e.g., 'LLaV A-KD', 'LLaV A-MoD', 'LLaV A-DI'); please standardize.","section":"§2.2","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"Please clarify whether the same epsilon is used in the logit normalization of Eq. (2) and in the arcosh clamp of Eq. (6), and state its numerical value for reproducibility.","section":"§3.2, Eqs. (2) and (6)","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The ablation means are over ten metrics with heterogeneous scales and are reported for seed 42 only; a per-seed variance for this ten-metric mean would let the reader compare the ablation gaps (0.0018-0.0081) with run-to-run noise directly.","section":"§4.4, Tables 11-16","authors_acknowledge":null,"authors_acknowledge_p":null}],"recommendation":"major_revision"},"author_rebuttal":null,"desk_editor":{"model":"deepseek/deepseek-v4-flash","letter":"Quick take: SurgPace is a well-built distillation framework, and the compressed 3B student results are plausible. But the paper's central empirical claim is not yet trustworthy, because the final configuration was chosen from ablation means computed on the fixed test split, and that same split is used for the headline 0.6955 average and the \"best on 15 of 16 metrics\" statement.\n\nThe genuinely new parts are worth naming. Separating sample-level curriculum from objective-level scheduling is a sensible design move. The min-fusion of teacher confidence and Lorentz dispersion as a difficulty score is a real novelty, even if the underlying ingredients (ULD-style cross-tokenizer logit alignment, LLaVA-KD-style hierarchical alignment, curriculum extraction) are known. The main comparison is also carefully controlled: same student architecture, same teacher, same initialization, same split, common generation protocol, and three-seed repeats for SurgPace versus EfficientVLM. That is more than many distillation papers do.\n\nThe soft spots are concentrated in evaluation. Tables 11 through 16 all report numbers on the test split, each marks the row that becomes the final configuration, and no validation-based selection is reported even though a 64,878-example validation split exists. So the headline is effectively the maximum of a model-selection loop over roughly twenty single-seed configurations evaluated on the test set. That inflates the margin over baselines and makes the \"best student method\" claim about a selected configuration, not an independent estimate. The concern is not hypothetical: the full-versus-no-curriculum gain in Table 14 is +0.0044 on ten ablated metrics, which is about the same size as the 16-metric seed standard deviation of 0.0038 in Table 4, and it is itself selected on test. So the central curriculum contribution, as measured, is weak evidence. I also notice some schedule constants (r_end, r1/r2/r3, start/end scales, temperature, gamma) are not reported, which makes reproduction harder, and the only curriculum baseline, CKD-MDSR, is out of domain and collapses.\n\nTo be fair, the paper acknowledges limitations around teacher dependence, fixed scores, hardware scope, and the LLM-based audit. It does not acknowledge the test-based configuration selection, and that is the load-bearing issue. The teacher and benchmark overlap with the authors' group is a smaller concern and not disqualifying on its own, especially because the teacher choice is reasonable and the efficiency table is concrete.\n\nThis paper deserves a serious referee. I would ask for a major revision: rerun selection on the validation split, report the hidden test numbers, release full hyperparameters and checkpoints, and add at least one in-domain curriculum baseline. If those numbers hold, this is a useful contribution for surgical MLLM compression. As it stands, the recipe is credible but the headline is not yet demonstrated.","headline":"A coherent distillation recipe and a real evaluation flaw: the headline numbers come from selecting the configuration on the same test split used to report them, so the central claim needs a validation-based rerun before it can be believed.","tokens_in":30814,"tokens_out":2000,"duration_ms":15022},"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek/deepseek-v4-flash","headline":"SurgPace claims that distilling a 13B surgical vision-language teacher into a 3B student works best as an annealed easy-to-hard curriculum, reporting the top student score on 15 of 16 benchmark metrics at about 4x lower memory use.","keywords":["knowledge distillation","multimodal large language models","curriculum learning","surgical data science","vision-language models","sample complexity estimation","hierarchical representation alignment","Lorentz geometry"],"falsifier":"Rank-correlate the teacher-derived score $s_i$ with the student's own per-sample loss (or per-sample loss reduction) at several checkpoints during distillation; if the correlation is near zero or negative, or if permuting the scores among samples of the same distribution leaves the 16-metric average unchanged, the complexity signal is not what drives the easy-to-hard gain. The complementary check the authors themselves flag as future work is distilling with a weaker or differently biased teacher and seeing whether the curriculum benefit survives.","tokens_in":31321,"feed_emoji":"🩺","tokens_out":14022,"duration_ms":84990,"temperature":0.7,"pith_summary":"SurgPace tries to establish that knowledge distillation of a large surgical multimodal language model into a small one works substantially better when it is run as a curriculum rather than as uniform transfer: samples are fed easy-first, and supervision shifts over time from imitating the teacher to optimizing against ground truth. Its central empirical claim is that a 3B student trained this way from a 13B teacher beats six established distillation baselines on 15 of 16 benchmark metrics, with an unweighted average of 0.6955 versus 0.6221 for the strongest baseline, while using about 4x less GPU memory and decoding about 1.4x faster. The stakes are clinical deployment: surgical assistants that answer questions, describe scenes, and localize instruments must run on operating-room hardware with strict latency limits, so a compression method that keeps capability while cutting cost is what the paper is after. The authors report the student even edges past the teacher on short-answer questions and LLM-judged description quality, while the teacher keeps its expected lead on open-ended and spatially demanding tasks.","feed_headline":"3B distilled surgical AI tops six baselines at 4x less memory","feed_subtitle":"An easy-to-hard schedule helps a 3B model beat six distillation baselines and near-match its 13B teacher.","key_machinery":"The load-bearing object is the offline complexity score $s_i = \\min(s_{\\mathrm{conf}}(x_i), s_{\\mathrm{lorentz}}(x_i))$, computed once from teacher decoding of the gold answer. $s_{\\mathrm{conf}}$ is $1$ minus the teacher's mean top-1 token probability; $s_{\\mathrm{lorentz}}$ lifts each teacher logit vector onto the unit Lorentz hyperboloid and measures the mean geodesic dispersion of the decoding trajectory, normalized across the whole dataset. The min-fusion is deliberately permissive: a sample counts as easy if either signal says so. The score becomes a sample weight $w'_i(p) = 1 + a(p)(w_i - 1)$ that anneals toward uniform weights as training progress $p$ reaches $r_{\\mathrm{end}}$, so training starts easy-first and ends balanced. Sample pacing is coupled to hierarchical representation alignment — temperature-scaled KL on the shared vocabulary support, squared distance between centered logits, latent feature alignment through a learned head, and entropy regularization — whose combined weight $\\beta_{\\mathrm{dist}}(p)$ is interpolated down while the cross-entropy weight $\\beta_{\\mathrm{ce}}(p)$ rises through a three-phase stage coefficient $\\alpha(p)$. Because teacher and student use different tokenizers, token positions are located by UTF-8 span and teacher logits are linearly interpolated at student positions before any comparison is made.","core_discovery":"The paper claims that distilling a 13B surgical multimodal teacher into a 3B student is markedly more effective when two training axes are coordinated: a sample-level curriculum that paces examples from easy to hard using an offline teacher-derived complexity score, and an objective-level schedule that shifts supervision from teacher imitation toward ground-truth task loss. On the Surg-396K benchmark, the produced student scores the highest unweighted average over 16 reported metrics (0.6955), is the best student method on 15 of 16 metrics, and repeats that dominance across three seeds ($0.6930\\pm0.0038$ versus $0.6221\\pm0.0029$ for the strongest baseline, EfficientVLM). A blinded LLM judge rates its detailed descriptions above the teacher's (77.48 versus 76.67), and controlled FP16 decoding shows $4.04\\times$ fewer parameters, about $4\\times$ less GPU memory, and $1.39\\times$ faster end-to-end latency than the 13B teacher. The authors state that the teacher still leads on open-answer overlap metrics, region-based reasoning, and grounding, which they attribute to a remaining capacity gap, and that their clinical audit is a reference-text LLM audit, not a substitute for prospective adjudication by surgeons looking at the original images.","pith_inferences":["The recipe is not surgical-specific: any heterogeneous vision-language corpus with a strong teacher could reuse the confidence-plus-Lorentz scorer, with pathology imaging, endoscopy, and procedural video as nearby testbeds.","The paper's own ablations put the curriculum's standalone gain at about 0.0044 on the ten-metric mean (0.6541 to 0.6585), so a fair reading is that the multi-level alignment and the progressive schedule carry most of the weight; a decisive test would be the full curriculum on top of a static objective.","Both difficulty signals are teacher-side and fixed offline, so the framework could be made adaptive by replacing the prior with a student-side measure such as per-sample student loss or teacher-student agreement — the open direction the authors list as future work.","The min-fusion rule is a falsifiable bet: a sample is called easy when only one of two teacher signals is confident, so samples where the teacher is confidently wrong will be mislabeled, which may be why the remaining audit errors concentrate in action direction and unsupported details."],"forward_implications":["A 3B surgical assistant can match or beat a 13B teacher on short-answer QA and LLM-judged description quality while cutting loaded GPU memory from 26.03 GB to 6.46 GB and end-to-end latency by about 1.4x in the paper's controlled FP16 benchmark.","Offline difficulty scoring is cheap and stable: scoring all 591,031 instances took 3.11 GPU-hours, and perturbing the teacher logits used for scoring costs only 0.4% (temperature $T=1.3$) to 1.0% (5% logit noise) on the 16-metric average.","The direction of the curriculum is what matters, not just difficulty weighting: easy-to-hard ordering beats hard-to-easy and mixed ordering on every reported metric, and linear pacing beats root, geometric, baby-step, and one-pass schedules.","The two design axes contribute separately: the full framework's ten-metric mean is 0.6585, versus 0.6541 with only the progressive objective schedule, and removing the latent-alignment or entropy terms degrades most metrics.","The hierarchical alignment objective is the strongest single component: removing either the KL or the centered-logit L2 term causes severe degradation across all task families."],"supporting_citations":[{"why":"Supplies the EndoChat 13B teacher, the Surg-396K benchmark with its five dialogue paradigms, and the LLM-as-a-Judge prompt reused for semantic evaluation.","marker":"[46]"},{"why":"Supplies the TinyLLaVA-style 3B student architecture and the pretrained Phi-2-SigLIP checkpoint that all compared methods share as initialization.","marker":"[18]"},{"why":"EfficientVLM is the strongest baseline; it is the comparison system for the three-seed robustness study and the blinded clinical audit.","marker":"[49]"},{"why":"MiniLLM is the standard LLM distillation baseline whose objective the student comparisons must beat.","marker":"[11]"},{"why":"ULD provides the cross-tokenizer logit distillation baseline, directly relevant since the teacher and student use different subword tokenizers.","marker":"[3]"},{"why":"LLaVA-KD is the multimodal distillation baseline focused on cross-modal knowledge transfer.","marker":"[4]"},{"why":"LLaVA-MoD is the MoE-based distillation baseline included in the student-method comparison.","marker":"[41]"},{"why":"CKD-MDSR is the curriculum-scheduled distillation baseline that SurgPace's sample-level curriculum directly contrasts with.","marker":"[43]"},{"why":"CAT-ViL contributes the EndoVis17/18-VQLA source data for the region-based and grounding QA tasks where the student shows its largest gains.","marker":"[1]"}],"fun_headline_variants":["Easy-to-hard curriculum helps 3B surgical model beat six distillation baselines","SurgPace distills a 13B surgical teacher into a 3B student that tops baselines","Paced distillation cuts surgical MLLM memory 4x while beating six baselines","Sample-level pacing plus shifting supervision lets 3B surgical model beat baselines","Teacher-guided easy-to-hard schedule yields compact surgical MLLM that beats baselines"],"cache_read_input_tokens":0,"weakest_assumption_plain":"The load-bearing premise is that a sample the 13B teacher decodes confidently, or with a compact logit trajectory, is genuinely easy for the 3B student to learn — and that the permissive min-fusion never lets one overconfident teacher signal mark a genuinely hard sample as easy; if teacher certainty does not transfer to student learnability, the curriculum signal is noise.","fun_headline_variants_meta":{"raw":{"variants":["Easy-to-hard curriculum helps 3B surgical model beat six distillation baselines","SurgPace distills a 13B surgical teacher into a 3B student that tops baselines","Paced distillation cuts surgical MLLM memory 4x while beating six baselines","Sample-level pacing plus shifting supervision lets 3B surgical model beat baselines","Teacher-guided easy-to-hard schedule yields compact surgical MLLM that beats baselines"]},"model":"deepseek/deepseek-v4-flash","effort":"low","cost_usd":0.00029,"raw_usage":{"cost":0.000290108,"is_byok":false,"cost_details":{"upstream_inference_cost":0.000290108,"upstream_inference_prompt_cost":3.8948e-05,"upstream_inference_completions_cost":0.00025116},"total_tokens":2568,"prompt_tokens":1188,"completion_tokens":1380,"prompt_tokens_details":{"audio_tokens":0,"video_tokens":0,"cached_tokens":950,"cache_write_tokens":0},"completion_tokens_details":{"audio_tokens":0,"image_tokens":0,"reasoning_tokens":1282}},"tokens_in":1188,"tokens_out":1380,"cost_source":"openrouter_usage","duration_ms":11737,"image_pages":[],"temperature":1.0,"billed_attempts":1,"reasoning_tokens":1282,"transport_provider":"openrouter","cache_read_input_tokens":0,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-11-03T17:37:18.720540+00:00","model_set":{"reader":"deepseek/deepseek-v4-flash"},"pipeline_version":"pith-pipeline@v1.4.1-daily-deepseek-or-medium","input_kind":null,"input_chars":null,"input_total_chars":null,"passes_missing":null,"published_standing":null,"published_confidence":null,"objection_audit":null,"run_record":null,"basis":{"kind":"not_recorded","source":null,"chars_given":null,"chars_total":null},"receipt":null,"falsifier":"Rank-correlate the teacher-derived score $s_i$ with the student's own per-sample loss (or per-sample loss reduction) at several checkpoints during distillation; if the correlation is near zero or negative, or if permuting the scores among samples of the same distribution leaves the 16-metric average unchanged, the complexity signal is not what drives the easy-to-hard gain. The complementary check the authors themselves flag as future work is distilling with a weaker or differently biased teacher and seeing whether the curriculum benefit survives.","supporting_citations":[{"cited_title":"The density-matrix renormalization group in the age of matrix product states,","cited_arxiv_id":null,"evidence_quote":"Supplies the EndoChat 13B teacher, the Surg-396K benchmark with its five dialogue paradigms, and the LLM-as-a-Judge prompt reused for semantic evaluation."},{"cited_title":"TinyLLaVA Factory: A Modularized Codebase for Small-scale Large Multimodal Models","cited_arxiv_id":"2405.11788","evidence_quote":"Supplies the TinyLLaVA-style 3B student architecture and the pretrained Phi-2-SigLIP checkpoint that all compared methods share as initialization."},{"cited_title":"EfficientVLM: Fast and Accurate Vision-Language Models via Knowledge Distillation and Modal-adaptive Pruning","cited_arxiv_id":null,"evidence_quote":"EfficientVLM is the strongest baseline; it is the comparison system for the three-seed robustness study and the blinded clinical audit."},{"cited_title":"MiniLLM: Knowledge distillation of large language models, in: The Twelfth International Conference on Learning Represen- tations","cited_arxiv_id":null,"evidence_quote":"MiniLLM is the standard LLM distillation baseline whose objective the student comparisons must beat."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"ULD provides the cross-tokenizer logit distillation baseline, directly relevant since the teacher and student use different subword tokenizers."},{"cited_title":"Llava-kd: A framework of distilling multimodal large language mod- els, in: Proceedings of the IEEE/CVF International Con- ference on Computer Vision, pp","cited_arxiv_id":null,"evidence_quote":"LLaVA-KD is the multimodal distillation baseline focused on cross-modal knowledge transfer."},{"cited_title":"General-purpose foundation models for increased autonomy in robot-assisted surgery","cited_arxiv_id":null,"evidence_quote":"LLaVA-MoD is the MoE-based distillation baseline included in the student-method comparison."},{"cited_title":"Curriculum-scheduled knowledge distillation from multiple pre-trained teachers for multi-domain se- quential recommendation","cited_arxiv_id":null,"evidence_quote":"CKD-MDSR is the curriculum-scheduled distillation baseline that SurgPace's sample-level curriculum directly contrasts with."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"CAT-ViL contributes the EndoVis17/18-VQLA source data for the region-based and grounding QA tasks where the student shows its largest gains."}],"lane_accuracy":{"status":"not yet measured","lane":{"model":"deepseek/deepseek-v4-flash","input_kind":null},"url":"https://pith.science/accuracy"},"review_version":1}