{"arxiv_id":"2607.11012","incoming":{"papers":1,"sentences":1,"groups":[],"unlabelled_sentences":1},"outgoing":{"papers":21,"sentences":31,"groups":[{"kind":"method_or_tool","label":"Uses the method or tool of","papers":1,"sentences":1,"quotes":[{"arxiv_id":"2409.19256","title":"HybridFlow: A Flexible and Efficient RLHF Framework","url":"/paper/2409.19256","section":"Experimental Setup","sentence":"All methods are implemented in EasyOPD and run on the same verl-based infrastructure [ref] .","intent_p":0.95,"contests_p":null}]},{"kind":"data_used","label":"Uses data or values from","papers":4,"sentences":5,"quotes":[{"arxiv_id":"2108.07732","title":"Program Synthesis with Large Language Models","url":"/paper/2108.07732","section":"Experimental Setup","sentence":"We evaluate on MATH500 [ref] , GSM8K [ref] , MBPP [ref] , and LiveCodeBench v6 [ref] , reporting exact-match accuracy for reasoning and pass@1 for code.","intent_p":0.93,"contests_p":null},{"arxiv_id":"2110.14168","title":"Training Verifiers to Solve Math Word Problems","url":"/paper/2110.14168","section":"Experimental Setup","sentence":"We evaluate on MATH500 [ref] , GSM8K [ref] , MBPP [ref] , and LiveCodeBench v6 [ref] , reporting exact-match accuracy for reasoning and pass@1 for code.","intent_p":0.93,"contests_p":null},{"arxiv_id":"2306.05301","title":"ToolAlpaca: Generalized Tool Learning for Language Models with 3000 Simulated Cases","url":"/paper/2306.05301","section":"Experimental Setup","sentence":"We evaluate Chemistry from SciKnowEval [ref] , Biology from SciKnowEval [ref] , Tool Use from ToolAlpaca [ref] , and mathematical reasoning on GSM8K [ref] .","intent_p":0.9,"contests_p":null},{"arxiv_id":"2406.09098","title":null,"url":"/paper/2406.09098","section":"Experimental Setup","sentence":"We evaluate Chemistry from SciKnowEval [ref] , Biology from SciKnowEval [ref] , Tool Use from ToolAlpaca [ref] , and mathematical reasoning on GSM8K [ref] .","intent_p":0.9,"contests_p":null}]},{"kind":"background","label":"Cites as background","papers":5,"sentences":5,"quotes":[{"arxiv_id":"2605.07711","title":"SimCT: Recovering Lost Supervision for Cross-Tokenizer On-Policy Distillation","url":"/paper/2605.07711","section":"Introduction","sentence":"OPD methods vary by supervision interface, tokenizer compatibility, and granularity, with overlapping settings including logit-based distillation, cross-tokenizer alignment [ref] , on-policy self-distillation, black-box or rubric-based feedback, and step-wise or multi-turn agentic supervision [ref] .","intent_p":0.81,"contests_p":null},{"arxiv_id":"2605.07725","title":"SOD: Step-wise On-policy Distillation for Small Language Model Agents","url":"/paper/2605.07725","section":"Introduction","sentence":"OPD methods vary by supervision interface, tokenizer compatibility, and granularity, with overlapping settings including logit-based distillation, cross-tokenizer alignment [ref] , on-policy self-distillation, black-box or rubric-based feedback, and step-wise or multi-turn agentic supervision [ref] .","intent_p":0.81,"contests_p":null},{"arxiv_id":"2606.15912","title":"On-Policy Distillation with Curriculum Turn-level Guidance for Multi-turn Agents","url":"/paper/2606.15912","section":"Introduction","sentence":"OPD methods vary by supervision interface, tokenizer compatibility, and granularity, with overlapping settings including logit-based distillation, cross-tokenizer alignment [ref] , on-policy self-distillation, black-box or rubric-based feedback, and step-wise or multi-turn agentic supervision [ref] .","intent_p":0.81,"contests_p":null},{"arxiv_id":"2407.01885","title":"Survey on Knowledge Distillation for Large Language Models: Methods, Evaluation, and Application","url":"/paper/2407.01885","section":"Introduction","sentence":"Conventional distillation uses teacher responses or token distributions on fixed contexts [ref] .","intent_p":0.94,"contests_p":null},{"arxiv_id":"1503.02531","title":"Distilling the Knowledge in a Neural Network","url":"/paper/1503.02531","section":"Background and Motivation","sentence":"Knowledge distillation (KD) typically trains a smaller student model to imitate the predictive behavior of a larger or more capable teacher [ref] .","intent_p":0.99,"contests_p":null}]}],"unlabelled_sentences":20},"claims":[],"thresholds":{"intent_p":0.8,"supports_p":0.95,"contests_p":0.7},"method":"Citing sentences come from each paper's LaTeX source; every \\cite marks the sentence and the works it cites. Pith's classifier typed each sentence. A label is shown at intent probability 0.8 or more (0.95 for support), a contest at 0.7 or more. Measured accuracy against two blind model families is at https://pith.science/accuracy. The quoted sentences are the citing authors' own.","held_up":null,"_links":{"html":"https://pith.science/paper/2607.11012"}}