| | | | Hello, curious Llamas! This is Amit Raja Naik, your friendly AI Human, writing for and from Sector 6. The launch of Llama 3 is creating quite a buzz! |
|
|
|
|
|
|
| Yesterday, Meta launched Llama 3, the most advanced open-source large language model, outperforming Claude 3 Sonnet and Google’s Gemini 1.5 Pro in reasoning and code generation capabilities. The excitement is nothing short of astounding. |
|
|
|
| | | | | Former OpenAI computer scientist Andrej Karpathy also expressed his excitement over the release of Llama 3, eagerly anticipating the capabilities of the forthcoming 400 billion model, which aims to compete with GPT-4. “400B is still training but already encroaching the GPT-4 territory,” he added, acknowledging the enhancements of released 8B and 70B models, currently available on GitHub and Hugging Face. However, he said that we will have to see when the rankings arrive on the LMSys Chatbot Arena Leaderboard. Interestingly, Llama 3 has already started climbing up the top of the leaderboard. It is currently in 5th position, surpassing Cohere's Command R+, Mistral Large and others. Further, he highlighted the benefits of increasing the token count in Llama 3. “With more tokens, you can compress sequences more in length, cite 15% fewer tokens, and see better downstream performance,” he added, saying how this leads to more efficient data processing and improved model performance. Limitations: While the increase in sequence length is a step in the right direction, Karpathy noted that it still falls short of the industry-leading standards. “The maximum number of tokens in the context window was bumped up to 8192... quite small w.r.t. modern standards," he added. Pushes the limits: However, Karpathy lauded the extensive training of the 8B model. “This is training ~75X beyond that point, which is unusual but personally, I think extremely welcome,” he added. Decoding Llama 3 architecture: “All models [now use] Grouped Query Attention (GQA), including the smallest 8B model,” revealed Karpathy, praising Llama 3’s architectural improvement, which extends across all sizes of the new model. |
|
|
|
| | | Meta drops Llama 3, but with a twist In a detailed Community License Agreement blog, Meta mentioned this under the redistribution and use section — “If you use the Llama materials to create, train, fine-tune, or otherwise improve an AI model, which is distributed or made available, you shall also include “Llama 3” at the beginning of any such AI model name.” Mixed reactions poured in from the Llama developer ecosystem over this. While some are open to the idea and see it as a responsible way to release a model, others question the reasoning behind such a forceful licensing clause. |
|
|
|
| | | | | | | Is Llama 3 good for Indic languages? Llama 3 is set to boost India's open AI ecosystem significantly; there's no doubt about that. However, a new conversation is brewing within the ecosystem regarding adapting it to Indic languages. He said that even though initial tests perform better with Devanagari than Llama 2, it struggles with other languages like Kannada and Tamil. More testing is needed to assess Llama 3's full potential with these languages. Further, he explained, in his blog that Llama3, using a TikToken-based tokeniser, struggles to tokenise Indic languages and may face difficulties expanding its vocabulary to accommodate them better. |
|
|
|
| The State of Data Engineering in India: 2024 |
|
|
|
| | | AIM Research recently released a detailed report on ‘The State of Data Engineering in India: 2024,’ highlighting rapid market growth alongside showcasing a significant rise in senior talent demand, salaries and more. Key Highlights:
- Market Growth: The data engineering market in India is expected to grow at a CAGR of 33.8%, increasing from $29.1 billion in 2023 to $124.7 billion by 2028.
- Senior Talent Demand: The proportion of data engineers with 6+ years of experience has risen significantly, from 27% in 2023 to 38% in 2024.
- Salary Growth: There has been a notable increase in the salary range, with the percentage of data engineers earning between INR 6-10 lakhs increasing from 24% in 2023 to 32.4% in 2024.
- Job Openings: The report found 10,593 job openings for data engineers across all industries on online job portals.
|
|
|
|
| | | OpenAI has hired Pragya Misra as its first employee in India to lead government relations as it seeks to navigate AI regulations and establish a local team amid the country's ongoing elections. |
|
|
|
| | | | | | | Join us for an exclusive webinar on ‘Automating Data Pipelines with Snowflake: Accelerating Data Integration and ETL Workflows’ on April 25, 2024. This insightful session is tailored for those keen on streamlining their data pipelines and accelerating time-to-insight. |
|
What are you waiting for? Register now to secure your spot in this enriching journey towards advanced data handling. Click here >> |
|
|
|
| Enjoying Sector 6 (formerly AIM Daily XO)? Share it with colleagues or friends – they can sign up here. We love hearing from our readers! Have thoughts on our new format? Questions, comments, or ideas are always welcome. If there’s a specific topic in AI or analytics that you're curious about, tell us! Reach out to us at info@analyticsindiamag.com. Stay tuned for more insights in our next edition!
Curated with ♥️ |
|
|
|
|
|
|
Комментариев нет:
Отправить комментарий
Примечание. Отправлять комментарии могут только участники этого блога.