Search for: All records

https://arxiv.org/pdf/2503.02951? 2025-acl-kodcode We introduce KodCode, a synthetic dataset that addresses the persistent challenge of acquiring high-quality, verifiable training data across diverse difficulties and domains for training Large Language Models for coding. Existing code-focused resources typically fail to ensure either the breadth of coverage (e.g., spanning simple coding tasks to advanced algorithmic problems) or verifiable correctness (e.g., unit tests). In contrast, KodCode comprises question-solution-test triplets that are systematically validated via a self-verification procedure. Our pipeline begins by synthesizing a broad range of coding questions, then generates solutions and test cases with additional attempts allocated to challenging problems. Finally, post-training data synthesis is done by rewriting questions into diverse formats and generating responses under a test-based reject sampling procedure from a reasoning model (DeepSeek R1). This pipeline yields a large-scale, robust and diverse coding dataset. KodCode is suitable for supervised fine-tuning and the paired unit tests also provide great potential for RL tuning. Fine-tuning experiments on coding benchmarks (HumanEval(+), MBPP(+), BigCodeBench, and LiveCodeBench) demonstrate that KodCode-tuned models achieve state-of-the-art performance, surpassing models like Qwen2.5-Coder-32B-Instruct and DeepSeek-R1-Distill-Llama-70B. 

The replication accuracy of DNA polymerase gamma (Pol γ) is essential for mitochondrial genome integrity. Mutation of human Pol γ arginine-853 has been linked to neurological diseases. Although not a catalytic residue, Pol γ arginine-853 mutants are void of polymerase activity. To identify the structural basis for the disease, we determined a crystal structure of the Pol γ mutant ternary complex with correct incoming nucleotide 2′-deoxycytidine 5′-triphosphate (dCTP). Opposite to the wild type that undergoes open-to-closed conformational changes when bound to a correct nucleotide that is essential for forming a catalytically competent active site, the mutant complex failed to undergo the conformational change, and the dCTP did not base pair with its Watson-Crick complementary templating residue. Our studies revealed that arginine-853 coordinates an interaction network that aligns the 3′-end of primer and dCTP with the catalytic residues. Disruption of the network precludes the formation of Watson-Crick base pairing and closing of the active site, resulting in an inactive polymerase. 

Abstract Extreme climate events are becoming more frequent, with poorly understood implications for carbon sequestration by terrestrial ecosystems. A better understanding will critically depend on accurate and precise quantification of ecosystems responses to these events. Taking the 2019 US Midwest floods as a case study, we investigate current capabilities for tracking regional flux anomalies with “top‐down” inversion analyses that assimilate atmospheric CO₂observations. For this analysis, we develop a regionally nested version of the NASA Carbon Monitoring System‐Flux system for North America (CMS‐Flux‐NA) that allows high resolution atmospheric transport (0.5° × 0.625°). Relative to a 2018 baseline, we find the 2019 US Midwest growing season net carbon uptake is reduced by 11–57 TgC (3%–16%, range across assimilated CO₂data sets). These estimates are found to be consistent with independent “bottom‐up” estimates of carbon uptake based on vegetation remote sensing (15–78 TgC). We then investigate current limitations in tracking regional carbon budgets using “top‐down” methods. In a set of observing system simulation experiments, we show that the ability of atmospheric CO₂inversions to capture regional carbon flux anomalies is still limited by observational coverage gaps for both in situ and satellite observations. Future space‐based missions that allow for daily observational coverage across North America would largely mitigate these observational gaps, allowing for improved top‐down estimates of ecosystem responses to extreme climate events. 

Abstract We select 48 multiflare gamma-ray bursts (GRBs) (including 137 flares) from the Swift/XRT database and estimate the spectral lag with the discrete correlation function. It is found that 89.8% of the flares have positive lags and only 9.5% of the flares show negative lags when fluctuations are taken into account. The median lag of the multiflares (2.75 s) is much greater than that of GRB pulses (0.18 s), which can be explained by the fact that we confirm that multiflare GRBs and multipulse GRBs have similar positive lag–duration correlations. We investigate the origin of the lags by checking the E peak evolution with the two brightest bursts and find the leading models cannot explain all of the multiflare lags and there may be other physical mechanisms. All of the results above reveal that X-ray flares have the same properties as GRB pulses, which further supports the observation that X-ray flares and GRB prompt-emission pulses have the same physical origin.