# Still Source: [Still]() Collection index: [llms.txt]() A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage. ## Benchmarks Page 1 of 1. This is an index page, not the full collection. - [Generalization Evaluation on AM-Thinking Distilled-math&code v1 (Unseen)](): Generalization Evaluation; AM-Thinking Distilled-math&code v1 (Unseen). - [Multi-view Subspace Clustering on Still DB](): Multi-view Subspace Clustering; Still DB. - [Safety Control on DeepSeek-R1-Distill-Qwen-1.5B](): Safety Control; DeepSeek-R1-Distill-Qwen-1.5B. - [Science Question Answering on DeepSeek-distilled ARC-C (test)](): Science Question Answering; DeepSeek-distilled ARC-C (test). - [Mathematical Reasoning on GSM8K DeepSeek-distilled (test)](): Mathematical Reasoning; GSM8K DeepSeek-distilled (test). - [License Plate Recognition on UCSD-Stills intra-dataset protocol (test)](): License Plate Recognition; UCSD-Stills intra-dataset protocol (test). - [Dialogue on Anthropic-HH (distillation set)](): Dialogue; Anthropic-HH (distillation set). - [Summarization on TL;DR (distillation set)](): Summarization; TL;DR (distillation set). - [Jailbreak Attack on JBDistill-Bench](): Jailbreak Attack; JBDistill-Bench. - [Efficiency Benchmarking on DeepSeek-R1-Distill-Llama-8B 16K Generation Length](): Efficiency Benchmarking; DeepSeek-R1-Distill-Llama-8B 16K Generation Length. - [Efficiency Benchmarking on DeepSeek-R1-Distill-Llama-8B (8K Generation Length)](): Efficiency Benchmarking; DeepSeek-R1-Distill-Llama-8B 8K Generation Length. - [AI-generated image detection on FakeForm Film Still](): AI-generated image detection; FakeForm Film Still. - [Video Generation on MAGI-1 4.5B-distill VBench gallery prompts 480x480 (test)](): Video Generation; MAGI-1 4.5B-distill VBench gallery prompts 480x480 (test). - [General Language Understanding on Downstream Benchmarks (Winogrande, HellaSwag, ARC-E, ARC-C, MMLU) (0.3B-token distillation)](): General Language Understanding; Downstream Benchmarks (Winogrande, HellaSwag, ARC-E, ARC-C, MMLU) 0.3B-token distillation. - [Distillation data detection on Distillation Data (test)](): Distillation data detection; Distillation Data (test). - [Knowledge Distillation Robustness on Llama-2-7B teacher vs. llama-2-7b-logit-watermark-distill-kgw-k1-gamma0.25-delta2 student (test)](): Knowledge Distillation Robustness; Llama-2-7B teacher vs. llama-2-7b-logit-watermark-distill-kgw-k1-gamma0.25-delta2 student (test). - [Knowledge Distillation Robustness on Llama-3.1-8B teacher vs. Llama-3.1-8B-Instruct-Open-R1-Distill student (test)](): Knowledge Distillation Robustness; Llama-3.1-8B teacher vs. Llama-3.1-8B-Instruct-Open-R1-Distill student (test). - [Knowledge Distillation Robustness on Qwen2.5-14B teacher vs. DeepSeek-R1-Distill-Qwen-14B student (test)](): Knowledge Distillation Robustness; Qwen2.5-14B teacher vs. DeepSeek-R1-Distill-Qwen-14B student (test). - [Knowledge Distillation Robustness on Qwen2.5-7B teacher vs. Qwen2.5-7B-Open-R1-Distill student (test)](): Knowledge Distillation Robustness; Qwen2.5-7B teacher vs. Qwen2.5-7B-Open-R1-Distill student (test). - [Membership Inference Attack on S1 Distillation Gemini-2.0-flash](): Membership Inference Attack; S1 Distillation Gemini-2.0-flash. - [Membership Inference Attack on S1.1 Distillation Deepseek-R1](): Membership Inference Attack; S1.1 Distillation Deepseek-R1. - [License Plate Recognition on UCSD-Stills (test)](): License Plate Recognition; UCSD-Stills (test). - [Multi-step LLM Interaction on DeepSeek-R1-Distill-Llama-8B Inference Performance (test)](): Multi-step LLM Interaction; DeepSeek-R1-Distill-Llama-8B Inference Performance (test). - [Speculative Decoding on Fineweb-edu distillation 8B to 300M](): Speculative Decoding; Fineweb-edu distillation 8B to 300M. - [Language Modeling on Fineweb-edu distillation 8B to 300M](): Language Modeling; Fineweb-edu distillation 8B to 300M. - [Style Manipulation Attack on DeepSeek-R1-Distill](): Style Manipulation Attack; DeepSeek-R1-Distill. - [Distillation Column State Prediction on Distillation Column Held-Out (test)](): Distillation column state prediction; Distillation column held-out (test). - [Video Generation on Wan2.1-1.3B 4-step distilled](): Video Generation; Wan2.1-1.3B 4-step distilled. - [LLM Attack Effectiveness on DeepSeek-R1-Distill-Llama-8B serving environment](): LLM Attack Effectiveness; DeepSeek-R1-Distill-Llama-8B serving environment. - [Adversarial Robustness on Jbdistill](): Adversarial Robustness; Jbdistill. - [Text-to-Video Generation on Wan2.1-14B-T2V-distilled (81 frames, 480P, 4steps, no cfg)](): Text-to-Video Generation; Wan2.1-14B-T2V-distilled 81 frames, 480P, 4steps, no cfg. - [Transfer Attack on Safety Neuron Transfer Attack Evaluation distilled vs base](): Transfer Attack; Safety Neuron Transfer Attack Evaluation distilled vs base. - [Cluster Validity Index Evaluation on Synthetic dataset DA2 closer clusters but still distinct](): Cluster Validity Index Evaluation; Synthetic dataset DA2 closer clusters but still distinct. - [Generalization gap prediction on Food101 Distillation](): Generalization gap prediction; Food101 Distillation. - [Novel Pose Synthesis Ablation on InterHand Distillation & Renderer 2.6M (test)](): Ablation Study (Novel Pose Synthesis); InterHand Distillation & Renderer Ablation 2.6M (test). - [End-to-end single-step decoding on DeepSeek-R1-Distill-LLaMA-8B 64K Context](): End-to-end single-step decoding; DeepSeek-R1-Distill-LLaMA-8B 64K Context. - [End-to-end single-step decoding on DeepSeek-R1-Distill-LLaMA-8B 32K Context](): End-to-end single-step decoding; DeepSeek-R1-Distill-LLaMA-8B 32K Context. - [End-to-end single-step decoding on DeepSeek-R1-Distill-LLaMA-8B 16K Context](): End-to-end single-step decoding; DeepSeek-R1-Distill-LLaMA-8B 16K Context. - [Speech Enhancement on StillSonicSet Noise-only condition](): Speech Enhancement; StillSonicSet Noise-only condition. - [Speech Enhancement on StillSonicSet Reverb+Noise](): Speech Enhancement; StillSonicSet Reverb+Noise. - [Speech Enhancement on StillSonicSet Reverb-only](): Speech Enhancement; StillSonicSet Reverb-only. - [Training Efficiency on Distillation (train)](): Training Efficiency; Distillation (train). - [Premise Generation on Rule Distillation Dataset LLM evaluation 1.0](): Premise Generation; Rule Distillation Dataset LLM evaluation 1.0. - [Premise Completion on Rule Distillation Dataset LLM evaluation 1.0](): Premise Completion; Rule Distillation Dataset LLM evaluation 1.0. - [Conclusion Generation on Rule Distillation Dataset LLM evaluation 1.0](): Conclusion Generation; Rule Distillation Dataset LLM evaluation 1.0. - [3D Face Alignment on Menpo-3D 68 landmarks (still frames)](): 3D Face Alignment; Menpo-3D 68 landmarks (still frames). - [Robotic Photography on 8 curated scenarios Still Life](): Robotic Photography; 8 curated scenarios Still Life. - [Value Modeling on DAPO-Math-17k DeepSeek-R1-Distill-Qwen-1.5B policy (Held-out)](): Value Modeling; DAPO-Math-17k DeepSeek-R1-Distill-Qwen-1.5B policy (Held-out). - [License Plate Recognition on UCSD-Stills](): License Plate Recognition; UCSD-Stills. - [Kicking in place on Still water water tank](): Kicking in place; Still water water tank. - [Clustering on Still](): Clustering; Still. - [On-Policy Distillation Feature Comparison on On-Policy Distillation Methods Feature Set](): On-Policy Distillation Feature Comparison; On-Policy Distillation Methods Feature Set.