Zephyr

Lewis Tunstall, E. Beeching, Nathan Lambert, Nazneen Rajani, Kashif Rasul, Younes Belkada, Shengyi Huang, Leandro von Werra, Clémentine Fourrier, Nathan Habib, Nathan Sarrazin, Omar Sanseviero, Alexander M. Rush, Thomas Wolf

We aim to produce a smaller language model that is aligned to user intent. Previous research has shown that applying distilled supervised fine-tuning (dSFT) on larger models significantly improves task accuracy; however, these models are unaligned, i.e. they do not respond well to natural prompts. To distill this property, we experiment with the use of preference data from AI Feedback (AIF). Starting from a dataset of outputs ranked by a teacher model, we apply distilled direct preference optimization (dDPO) to learn a chat model with significantly improved intent alignment. The approach requires only a few hours of training without any additional sampling during fine-tuning. The final result, Zephyr-7B, sets the state-of-the-art on chat benchmarks for 7B parameter models, and requires no human annotation. In particular, results on MT-Bench show that Zephyr-7B surpasses Llama2-Chat-70B, the best open-access RLHF-based model. Code, models, data, and tutorials for the system are available at https://github.com/huggingface/alignment-handbook.

Yihao Wu, Jooyoung Bae, Chaeyun Shim, Bongjin Kim

T. Koster, P. Shah, A. Valipour, G. Criner, F. Herth, R. Sue, D. K. Hogarth, Ralitza T. Martin, Amit K. Mahajan, R. Alalawi, Lisa Kopas, Avi Cohen, D. Wood, J. Kurman, N. Shargill, M. Dransfield, D. Slebos, Michael Perch

Bronchoscopic lung volume reduction treatment with Zephyr one-way valves is an effective guideline-based treatment option for patients with severe emphysema and hyperinflation. However, in some cases the treatment response is less than anticipated or there might be a loss of initial treatment effect. Reasons for the lack of response can include incorrect assessment of collateral ventilation, improper valve placement, or patient related factors. Loss of initial benefit can be due to granulation tissue formation and subsequent valve dysfunction, or there may be side effects such as excessive coughing or infectious problems. Careful follow-up after treatment with valves is important and evaluation with a CT scan and/or bronchoscopy is helpful if there is no improvement after treatment or loss of initial benefit. This paper aims to describe the most important causes and provide a strategy of how to approach and manage these patients.

J. Hatzl, M. Van Basten Batenburg, K. Yeung, B. Fioole, Eric L. G. Verhoeven, G. Lauwers, T. Kölbel, J. Wever, Dierk Scheinert, W. van den Eynde, G. Rouhani, B. Mees, F. Vermassen, H. Schelzig, D. Böckler, P. Cuypers

OBJECTIVE The midterm outcomes of the low profile Zenith Alpha Abdominal Endovascular Graft from the ZEnith alPHa for aneurYsm Repair (ZEPHYR) registry are reported. METHODS The ZEPHYR registry is a physician initiated, multicentre, non-randomised, core laboratory controlled, prospective registry. Inclusion criteria were patients with non-ruptured abdominal aortic aneurysm with a maximum diameter ≥ 50 mm or enlargement > 5 mm within 6 months, with a site reported infrarenal neck length of ≥ 10 mm and with the intention to electively implant the Zenith Alpha abdominal endograft. Patients were included in 14 sites across Germany, Belgium, and the Netherlands. The primary endpoint was treatment success, defined as technical success and clinical success. Technical success was defined as successful delivery and deployment of the endograft in the planned position without unintentional coverage of internal iliac or renal arteries, with successful removal of the delivery system. Clinical success was defined as freedom from aneurysm sac expansion > 5 mm, type I or type III endoleaks, aneurysm rupture, stent graft migration > 10 mm, open conversion, and stent graft occlusion. RESULTS A total of 347 patients were included in the ZEPHYR registry. Median clinical follow up was 743 days (interquartile range [IQR] 657, 806) with a median imaging follow up of 725 days (IQR 408, 788). Treatment success at 6 months, 1 year, and 2 years was 92.5%, 90.4%, and 85.3%, respectively. Freedom from secondary intervention was 94.3%, 93.4%, and 86.9%, respectively. The predominant cause for secondary interventions was limb complications. Freedom from limb occlusion (per patient) at 6 months, 1 year, and 2 years was 97.2%, 95.8%, and 92.5%, respectively. Univariate and multivariate Cox regression analyses could not identify any independent predictor for limb complications. CONCLUSION While treatment success is comparable with other commercially available grafts, the rate of limb complications at 2 years is concerning. The manufacturer's instructions for use should be closely followed. Further studies are necessary to investigate the root cause of the increased rate of limb complications with the Zenith Alpha Abdominal Endovascular Graft.

Yub Raj Sehdai, Hassnain Awan, P. Roy, Faisal Chowdhury, A. Siddiq, S. Shrestha, Harendra Singh, Sanskriti Chapagain, S. Khanal, S. Kadariya, S. Rayamajhi, Aimi Shreshtha, Aayush Shrestha, M. Ahmed, Nisarfathima Kazimuddin, T. Khan, Karan Singh, Irfan Waheed, R. Steff

Background: Chronic obstructive pulmonary disease (COPD) causes permanent lung damage and is a global health issue. Bronchoscopic Lung Volume Reduction (BLVR) with Zephyr valves has increased COPD treatment choices for individuals with persistent symptoms despite maximal drug therapy. Objectives: This systematic review and meta-analysis evaluated available Zephyr valve data in severe emphysema patients. Data and methods: We searched Cochrane, PubMed, Google Scholar, Lilacs, and Science Direct extensively and included eight randomized control trials (RCTs). Results: This meta-analysis comprises 1083 patients 669 Zephyr valve-treated and 414 control. The mean difference in 6-minute walking distance was 37.53 meters (95% CI: 22.14–52.90, P < 0.001). Statistically significant improvements were also observed in FEV1 predicted percentage (mean difference: 13.64, 95% CI: 8.51–18.77, P < 0.001) and BODE index change (mean difference: −0.79, 95% CI: −1.33 to 0.26, P < 0.05), mMRC dyspnea scale (mean difference: −0.27, 95% CI: −0.55 to 0.02, P < 0.01) and residual volume (mean difference: −0.14, 95% CI: −0.64 to 0.35, P < 0.01). The quality of life measured by SGRQ also improved significantly (mean difference: −6.69, 95% CI: −9.69, −4.23, P < 0.001). While COPD exacerbation rates (OR:1.41, 95% CI: 0.87–2.30, P > 0.05) and pneumonia rates (OR: 1.72, 95% CI: 0.7–4.23, P > 0.05) were similar between valve and control groups, pneumothorax risk was significantly higher in the treatment arm (OR: 11.06, 95% CI: 4.10–29.80, P < 0.001). However, there was no significant difference in all-cause mortality between the groups (OR: 1.69; 95% CI: 0.53–5.38, P > 0.38). Conclusion: Zephyr valves may improve lung function, exercise capacity, and quality of life in patients suffering from severe emphysema. However, the increased risk of pneumothorax requires careful patient selection and rigorous post-intervention monitoring.