HE-SNR is a high-entropy signal-to-noise ratio metric derived from the Entropy Compression Hypothesis to better guide LLM mid-training on complex software engineering benchmarks.
forking points
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
HE-SNR: Uncovering Latent Logic via Entropy for Guiding Mid-Training on SWE-bench
HE-SNR is a high-entropy signal-to-noise ratio metric derived from the Entropy Compression Hypothesis to better guide LLM mid-training on complex software engineering benchmarks.