Workflow
COMET
icon
Search documents
X @Bitget
Bitget· 2025-08-02 05:39
New Listing & Cryptocurrency - Bitget Onchain 上线新的加密货币项目,寻找下一个百倍币 [1] - 新上线的加密货币包括 $ZEUS, BONKPUTER, $BSTR, Bubble, $JUGS, $COMET, $AiMau [1] Platform Information - 用户可以通过 Bitget App 的 Markets → Onchain 访问这些新币 [1]
TECH Gears Up to Report Q3 Earnings: Here's What to Expect
ZACKS· 2025-04-30 12:55
Core Viewpoint - Bio-Techne Corporation (TECH) is expected to report its third-quarter fiscal 2025 results on May 7, with positive earnings and revenue growth anticipated based on recent performance and market trends [1][2]. Group 1: Financial Performance Estimates - The Zacks Consensus Estimate for Bio-Techne's revenues is $315.2 million, reflecting a 3.9% increase from the previous year [2]. - The consensus estimate for EPS is projected at 51 cents, indicating a 6.3% year-over-year improvement [2]. - Recent estimates for EPS have decreased by 1.9% over the past 30 days [2]. Group 2: Segment Performance Insights - The Protein Sciences segment is expected to benefit from its protein analysis portfolio and GMP protein business, with revenues estimated at $224.7 million, up 4.7% from the prior year [3][5]. - The Diagnostics and Spatial Biology segment is anticipated to show solid growth, particularly in the spatial biology vertical, with revenues estimated at $92.5 million, a 5.7% increase from the previous year [6][9]. Group 3: Recent Developments and Innovations - Bio-Techne launched its Simple Western platform called Leo, which allows for high-throughput analysis of samples [4]. - The company expanded its geographic reach through new distribution agreements in Turkey and Azerbaijan, which may positively impact revenue [4]. - The Precision Diagnostic Tools business is experiencing growth, with a unique portfolio that continues to gain market share [7]. Group 4: Collaborations and Product Launches - Bio-Techne's Asuragen brand collaborated with Oxford Nanopore Technologies to launch a new genetic panel, which is expected to enhance business performance [9]. - The company also expanded its menu of RNAscope in situ hybridization probes to support spatial biology research and next-generation therapeutics [8].
富士康弄了个大模型;零跑提前季度盈利;Cursor母公司寻求百亿美元估值丨百亿美元公司动向
晚点LatePost· 2025-03-11 03:06
富士康也弄了个大模型。 富士康的大模型叫 FoxBrain,以 Meta Llama 3.1 为基础架构,具备推理能力。富士康说他们用了 120 块英伟达 H100 在大约四周内完成了 FoxBrain 的训练。他们打算将 FoxBrain 开源,以便与行 业伙伴合作。 零跑汽车营收同比接近翻倍,提前实现季度盈利。 2024 年,零跑汽车共交付了 29.27 万台新车,同比增长 103.8%,带动营收同比增长 92% 至 321.6 亿元,毛利率从去年的 0.5% 改善至 8.4%。四季度,零跑首次实现单季度净利润、较原计划提前一 年,规模约为 8000 万元。 零跑的销售主力为 C 系列,C 系列销量占去年总销量的 76.6%。截至去年末,零跑共有销售门店 695 家,服务门店 427 家。零跑表示,通过优化经销商结构和投资人区域分布的策略,去年渠道盈 利率超过 80%。 引发全球 AI 编程热的 Cursor 母公司寻求百亿美元估值。 媒体报道,开发 AI 编程助手 Cursor 的母公司 Anysphere 正在以估值 100 亿美元寻求融资。三个月 前,Anysphere 的估值是 25 亿美元。上一 ...
字节跳动,重大宣布!成本再降40%!
证券时报· 2025-03-10 12:43
Core Viewpoint - The article discusses the significant advancements in large model training efficiency and cost reduction through the introduction of the COMET optimization technology by ByteDance's Doubao model team, which enhances training efficiency by 1.7 times and reduces costs by 40% [1][2]. Group 1: COMET Technology - COMET is a key optimization technology for the MoE (Mixture of Experts) architecture, which has been open-sourced on GitHub, leading to substantial savings in GPU hours during training [1]. - The technology addresses the communication overhead in distributed training of MoE models, significantly improving training efficiency and cost [1][2]. - COMET can be integrated into existing MoE training frameworks without invasive modifications, making it more flexible and widely applicable [2]. Group 2: Synergy with Other Technologies - COMET can be used in conjunction with the previously introduced DualPipe+DeepEP solution, allowing for further cost reductions in training [2]. - The UltraMem sparse model architecture, which was announced earlier, can also be combined with COMET to enhance training efficiency and reduce costs even further [2]. Group 3: Market Trends and Cost Reduction - The cost of calling large models has drastically decreased, with estimates suggesting a drop from $120 per million tokens in 2023 to less than 1 RMB in 2024, representing a 99.9% reduction [3]. - The reduction in training costs is expected to lower the barriers for enterprises, potentially increasing demand for larger and more complex models [3]. Group 4: Investment in AI Talent and Infrastructure - ByteDance is significantly increasing its capital expenditure in AI, with projections of reaching 80 billion RMB in 2024, nearly matching the combined total of major competitors [4]. - The company is actively recruiting top AI talent through initiatives like the "Jindouyun Talent Program" and the "Top Seed Talent Program," focusing on attracting students interested in AI research [5].