Search for: All records

The BigCode community, an open-scientific collaboration working on the responsible development of Large Language Models for Code (Code LLMs), introduces StarCoder and StarCoderBase: 15.5B parameter models with 8K context length, infilling capabilities and fast large-batch inference enabled by multi-query attention. StarCoderBase is trained on 1 trillion tokens sourced from The Stack, a large collection of permissively licensed GitHub repositories with inspection tools and an opt-out process. We fine-tuned StarCoderBase on 35B Python tokens, resulting in the creation of StarCoder. We perform the most comprehensive evaluation of Code LLMs to date and show that StarCoderBase outperforms every open Code LLM that supports multiple programming languages and matches or outperforms the OpenAI code-cushman-001 model. Furthermore, StarCoder outperforms every model that is fine-tuned on Python, can be prompted to achieve 40% pass@1 on HumanEval, and still retains its performance on other programming languages. We take several important steps towards a safe open-access model release, including an improved PII redaction pipeline and a novel attribution tracing tool, and make the StarCoder models publicly available under a more commercially viable version of the Open Responsible AI Model license. 

Xenon dual-phase time projections chambers (TPCs) have proven to be a successful technology in studying physical phenomena that require low-background conditions. With$$40\,\textrm{t}$$$40\text{t}$of liquid xenon (LXe) in the TPC baseline design, DARWIN will have a high sensitivity for the detection of particle dark matter, neutrinoless double beta decay ($$0\upnu \upbeta \upbeta $$$0\nu \beta \beta$), and axion-like particles (ALPs). Although cosmic muons are a source of background that cannot be entirely eliminated, they may be greatly diminished by placing the detector deep underground. In this study, we used Monte Carlo simulations to model the cosmogenic background expected for the DARWIN observatory at four underground laboratories: Laboratori Nazionali del Gran Sasso (LNGS), Sanford Underground Research Facility (SURF), Laboratoire Souterrain de Modane (LSM) and SNOLAB. We present here the results of simulations performed to determine the production rate of$${}^{137}$$${}^{137}$Xe, the most crucial isotope in the search for$$0\upnu \upbeta \upbeta $$$0\nu \beta \beta$of$${}^{136}$$${}^{136}$Xe. Additionally, we explore the contribution that other muon-induced spallation products, such as other unstable xenon isotopes and tritium, may have on the cosmogenic background.

 

Biological organisms are increasingly being introduced and eradicated in an effort to maintain biodiversity and ecosystem function in the face of anthropogenic threats. However, these conservation actions can have unintended consequences to non‐target species. Careful vetting of these actions using ecological modelling tools could help predict and avoid unintended consequences.

Qualitative modelling tools, such as fuzzy interaction webs (FIWs), allow for qualitative rankings of community properties (e.g. interaction strength = high, medium, low) in combination with quantitative information to predict management outcomes. These tools have lower data requirements than strictly quantitative models, facilitating their use for communities lacking comprehensive parameterization. However, no studies have evaluated the efficacy of FIWs for predicting unintended consequences against empirically documented outcomes. Moreover, there is no process for systematically identifying which species to incorporate in community‐level conservation assessments to overcome model structure uncertainty. Finally, there is a need to make qualitative modelling tools more accessible for conservation practitioners.

We applied FIWs to the case study of lake trout introduction into Yellowstone Lake, Yellowstone National Park, to assess its ability to predict documented community‐level outcomes from an intentional species introduction. Next, we used the case study of the intentional red squirrel introduction to Newfoundland to show how a community assessment framework can help define the community interaction web needed for applying a FIW. Lastly, we introduced a user‐friendly web interface (https://matrix.mpgranch.com/#/) for applying FIWs to conservation questions.

We found that the FIW predicted previously documented directional changes in the abundance of community components relatively well in the Yellowstone Lake case study, even with minimal knowledge of the system. The community assessment framework provided a formal process for identifying community components for the Newfoundland case study, and the resulting FIW predicted documented unintended consequences. The user interface predicts realistic outcomes in our study system and allows managers to build and apply FIWs for conservation planning.

Synthesis and applications. Our community assessment framework and user interface can be used to apply FIWs to identify and avert potential unintended outcomes of species introductions and eradications for improved conservation management.

 

Abstract The XENONnT detector uses the latest and largest liquid xenon-based time projection chamber (TPC) operated by the XENON Collaboration, aimed at detecting Weakly Interacting Massive Particles and conducting other rare event searches.The XENONnT data acquisition (DAQ) system constitutes an upgraded and expanded version of the XENON1T DAQ system.For its operation, it relies predominantly on commercially available hardware accompanied by open-source and custom-developed software.The three constituent subsystems of the XENONnT detector, the TPC (main detector), muon veto, and the newly introduced neutron veto, are integrated into a single DAQ, and can be operated both independently and as a unified system.In total, the DAQ digitizes the signals of 698 photomultiplier tubes (PMTs), of which 253 from the top PMT array of the TPC are digitized twice, at ×10 and ×0.5 gain.The DAQ for the most part is a triggerless system, reading out and storing every signal that exceeds the digitization thresholds.Custom-developed software is used to process the acquired data, making it available within ∼30 s for live data quality monitoring and online analyses.The entire system with all the three subsystems was successfully commissioned and has been operating continuously, comfortably withstanding readout rates that exceed ∼500 MB/s during calibration.Livetime during normal operation exceeds 99% and is ∼90% during most high-rate calibrations.The combined DAQ system has collected more than 2 PB of both calibration and science data during the commissioning of XENONnT and the first science run. 

Abstract A low-energy electronic recoil calibration of XENON1T, a dual-phase xenon time projection chamber, with an internal $${}^{37}$$ 37 Ar source was performed. This calibration source features a 35-day half-life and provides two mono-energetic lines at 2.82 keV and 0.27 keV. The photon yield and electron yield at 2.82 keV are measured to be ( $$32.3\,\pm \,0.3$$ 32.3 ± 0.3 ) photons/keV and ( $$40.6\,\pm \,0.5$$ 40.6 ± 0.5 ) electrons/keV, respectively, in agreement with other measurements and with NEST predictions. The electron yield at 0.27 keV is also measured and it is ( $$68.0^{+6.3}_{-3.7}$$ 68 . 0 - 3.7 + 6.3 ) electrons/keV. The $${}^{37}$$ 37 Ar calibration confirms that the detector is well-understood in the energy region close to the detection threshold, with the 2.82 keV line reconstructed at ( $$2.83\,\pm \,0.02$$ 2.83 ± 0.02 ) keV, which further validates the model used to interpret the low-energy electronic recoil excess previously reported by XENON1T. The ability to efficiently remove argon with cryogenic distillation after the calibration proves that $${}^{37}$$ 37 Ar can be considered as a regular calibration source for multi-tonne xenon detectors. 

Abstract The XENON collaboration has published stringent limits on specific dark matter – nucleon recoil spectra from dark matter recoiling on the liquid xenon detector target. In this paper, we present an approximate likelihood for the XENON1T 1 t-year nuclear recoil search applicable to any nuclear recoil spectrum. Alongside this paper, we publish data and code to compute upper limits using the method we present. The approximate likelihood is constructed in bins of reconstructed energy, profiled along the signal expectation in each bin. This approach can be used to compute an approximate likelihood and therefore most statistical results for any nuclear recoil spectrum. Computing approximate results with this method is approximately three orders of magnitude faster than the likelihood used in the original publications of XENON1T, where limits were set for specific families of recoil spectra. Using this same method, we include toy Monte Carlo simulation-derived binwise likelihoods for the upcoming XENONnT experiment that can similarly be used to assess the sensitivity to arbitrary nuclear recoil signatures in its eventual 20 t-year exposure.