A per-token feature from temperature-induced changes in LLM token distributions predicts within-prompt creativity rank at Spearman rho 0.918 vs LLM judges and 0.870 vs humans, outperforming perplexity, entropy, top-1 margin, and compression baselines.
Texygen: A Benchmarking Platform for Text Generation Models
5 Pith papers cite this work, alongside 156 external citations. Polarity classification is still indexing.
5
Pith papers citing it
156
external citations · Pith
abstract
We introduce Texygen, a benchmarking platform to support research on open-domain text generation models. Texygen has not only implemented a majority of text generation models, but also covered a set of metrics that evaluate the diversity, the quality and the consistency of the generated texts. The Texygen platform could help standardize the research on text generation and facilitate the sharing of fine-tuned open-source implementations among researchers for their work. As a consequence, this would help in improving the reproductivity and reliability of future research work in text generation.
verdicts
UNVERDICTED 5representative citing papers
Introduces TBPO, which derives a Bregman-divergence density-ratio matching objective for token-level preference optimization that generalizes DPO while preserving the induced optimal policy.
Decan (D_Ca_n = C × a_n) measures text diversity as progressive conditional surprise from base LM log-probabilities, scoring 0.846 OCA on McDiv benchmark and detecting monotonic diversity drop across base→SFT→DPO→RLVR stages.
TOFU loss mitigates the narrowing of generative diversity in LLMs after supervised fine-tuning by addressing neglect of low-frequency patterns and forgetting of prior knowledge.
Only gated RNN language models reproduce the long-range correlation scaling of natural language among tested models, with Taylor's law exponent serving as a quality indicator.
citing papers explorer
-
Before and After Temperature: A Distributional View of Creative LLM Generation
A per-token feature from temperature-induced changes in LLM token distributions predicts within-prompt creativity rank at Spearman rho 0.918 vs LLM judges and 0.870 vs humans, outperforming perplexity, entropy, top-1 margin, and compression baselines.
-
TokenRatio: Principled Token-Level Preference Optimization via Ratio Matching
Introduces TBPO, which derives a Bregman-divergence density-ratio matching objective for token-level preference optimization that generalizes DPO while preserving the induced optimal policy.
-
"I've Seen How This Goes": Characterizing Diversity via Progressive Conditional Surprise
Decan (D_Ca_n = C × a_n) measures text diversity as progressive conditional surprise from base LM log-probabilities, scoring 0.846 OCA on McDiv benchmark and detecting monotonic diversity drop across base→SFT→DPO→RLVR stages.
-
Diversity in Large Language Models under Supervised Fine-Tuning
TOFU loss mitigates the narrowing of generative diversity in LLMs after supervised fine-tuning by addressing neglect of low-frequency patterns and forgetting of prior knowledge.
-
Evaluating Computational Language Models with Scaling Properties of Natural Language
Only gated RNN language models reproduce the long-range correlation scaling of natural language among tested models, with Taylor's law exponent serving as a quality indicator.