HE-SNR: Uncovering Latent Logic via Entropy for Guiding Mid-Training on SWE-bench

Yueyang Wang, Jiawei Fu, Baolong Bi, Xili Wang, Xiaoqing Liu
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:126195-126217, 2026.

Abstract

SWE-bench has emerged as the premier benchmark for evaluating Large Language Models on complex software engineering tasks. While these capabilities are fundamentally acquired during the mid-training phase and subsequently elicited during Supervised Fine-Tuning (SFT), there remains a critical deficit in metrics capable of guiding mid-training effectively. Standard metrics such as Perplexity (PPL) are compromised by the "Long-Context Tax" and exhibit weak correlation with downstream SWE performance. In this paper, we bridge this gap by first introducing a rigorous data filtering strategy. Crucially, we propose the Entropy Compression Hypothesis, redefining intelligence not by scalar Top-1 compression, but by the capacity to structure uncertainty into Entropy-Compressed States of low orders ("reasonable hesitation"). Grounded in this fine-grained entropy analysis, we formulate a novel metric, HE-SNR (High-Entropy Signal-to-Noise Ratio). We validate our approach on models with up to 560B parameters across different context windows (32K/128K). This work provides both the theoretical foundation and practical tools for optimizing the latent potential of LLMs in complex engineering domains.

Cite this Paper


BibTeX
@InProceedings{pmlr-v306-wang26bd, title = {{HE}-{SNR}: Uncovering Latent Logic via Entropy for Guiding Mid-Training on {SWE}-bench}, author = {Wang, Yueyang and Fu, Jiawei and Bi, Baolong and Wang, Xili and Liu, Xiaoqing}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {126195--126217}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/wang26bd/wang26bd.pdf}, url = {https://proceedings.mlr.press/v306/wang26bd.html}, abstract = {SWE-bench has emerged as the premier benchmark for evaluating Large Language Models on complex software engineering tasks. While these capabilities are fundamentally acquired during the mid-training phase and subsequently elicited during Supervised Fine-Tuning (SFT), there remains a critical deficit in metrics capable of guiding mid-training effectively. Standard metrics such as Perplexity (PPL) are compromised by the "Long-Context Tax" and exhibit weak correlation with downstream SWE performance. In this paper, we bridge this gap by first introducing a rigorous data filtering strategy. Crucially, we propose the Entropy Compression Hypothesis, redefining intelligence not by scalar Top-1 compression, but by the capacity to structure uncertainty into Entropy-Compressed States of low orders ("reasonable hesitation"). Grounded in this fine-grained entropy analysis, we formulate a novel metric, HE-SNR (High-Entropy Signal-to-Noise Ratio). We validate our approach on models with up to 560B parameters across different context windows (32K/128K). This work provides both the theoretical foundation and practical tools for optimizing the latent potential of LLMs in complex engineering domains.} }
Endnote
%0 Conference Paper %T HE-SNR: Uncovering Latent Logic via Entropy for Guiding Mid-Training on SWE-bench %A Yueyang Wang %A Jiawei Fu %A Baolong Bi %A Xili Wang %A Xiaoqing Liu %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-wang26bd %I PMLR %P 126195--126217 %U https://proceedings.mlr.press/v306/wang26bd.html %V 306 %X SWE-bench has emerged as the premier benchmark for evaluating Large Language Models on complex software engineering tasks. While these capabilities are fundamentally acquired during the mid-training phase and subsequently elicited during Supervised Fine-Tuning (SFT), there remains a critical deficit in metrics capable of guiding mid-training effectively. Standard metrics such as Perplexity (PPL) are compromised by the "Long-Context Tax" and exhibit weak correlation with downstream SWE performance. In this paper, we bridge this gap by first introducing a rigorous data filtering strategy. Crucially, we propose the Entropy Compression Hypothesis, redefining intelligence not by scalar Top-1 compression, but by the capacity to structure uncertainty into Entropy-Compressed States of low orders ("reasonable hesitation"). Grounded in this fine-grained entropy analysis, we formulate a novel metric, HE-SNR (High-Entropy Signal-to-Noise Ratio). We validate our approach on models with up to 560B parameters across different context windows (32K/128K). This work provides both the theoretical foundation and practical tools for optimizing the latent potential of LLMs in complex engineering domains.
APA
Wang, Y., Fu, J., Bi, B., Wang, X. & Liu, X.. (2026). HE-SNR: Uncovering Latent Logic via Entropy for Guiding Mid-Training on SWE-bench. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:126195-126217 Available from https://proceedings.mlr.press/v306/wang26bd.html.

Related Material