Waii

A Quick Deepseek R1 Testing for SQL Generation (with Waii) Finally got to test Deepseek R1! Tried both versions: distilled LLaMA 8B from R1 (runs in local ollama) and Deepseek R1 (671B) from fireworks. Here's what I discovered: 1) The self-debate is fascinating - it's like watching a real, capable but hesitant person think through problems. Check out this snippet for one of the query: --- Okay, let's tackle this query. The user wants countries that have consistently below median GDP per capita but above median GDP from 2018 to 2022. First, I need to identify the correct indicators. The WDI table has "Indicator Code" values. From the provided column values, 'NY.GDP.PCAP.KD' is GDP per capita (constant 2015 US$), and 'NY.GDP.MKTP.KD' is GDP (constant 2015 US$). These seem like the right indicators to use. Next, since the WDI table is a slowly changing dimension, but the user specifies a time range (2018-2022), we don't need to filter for the most recent year. Instead, we'll use the years 2018 to 2022 directly. We need to calculate the median for each indicator across all countries for each year in that range. Then, compare each country's yearly values against these medians. The country must have all years below the GDP per capita median and above the GDP median. ... --- 2) R1 is REALLY good at understanding aggregation. Current models like GPT-4o, Claude 3.5 sometimes mess up complex aggregations (like avg of daily avg sales) or window functions with filters. But R1 handles these consistently better than other models I've tested. Here's a perfect example - a query to "find countries that moved up in GDP rankings by at least 2 positions each year from 2018 to 2022". You can read the attached SQL query snippet, notice how clean the window functions and filters are. This is much better / consistent than gpt-4o / sonnet 3.5. Downsides? Everything comes with a cost: - Model takes long <think> output for any task, even simple entity extraction - I still need to use GPT-4o for quick tasks (reranking, entity extraction) during the test, otherwise it will take forever. - Query generation takes 4-5 mins vs 10-20 secs with GPT-4o - Most time is spent on unnecessary self-debate. The solution is clear in first 20-30% (40-60 secs), but it keeps rewriting and second-guessing the correct solution About the distilled model - don't get too excited. It can't handle complexity: - Struggles with AMC-8 (8th-grade math competition) questions -- just went into an infinite loop of output and cannot give me an answer. - Can't generate SQL based on the schema input. - Only useful for simple tasks that basic models already handle well. Looking forward: While these new reasoning models (O3, Gemini-2.0-thinking) aren't fully practical yet, they show huge potential. Once they solve the fast/slow thinking problem and learn when to stop deliberating, they'll be game-changing. My prediction? The future belongs to such thinking models! #deepseek-r1, #deepseek, #text2sql

Annoyed you can only do Text-to-SQL on top of one table and even then your answers are wrong and full of LLM hallucinations? See how some of the largest companies in the world today are using Waii.ai's most accurate Text-to-SQL across their 1000s/10,000s of tables for true enterprise natural language analytics. Register below for this free webinar led by ex-{Snowflake, Cloudera, and Amazon} engineering and product leaders.

Trino has all the data accessible. Waii has the tools to convert your text to SQL, charts and more. And now waii supports Trino - check?https://lnkd.in/gRGGvN_N?for more info and give it a whirl. Thanks to Gunther Hagleitner and Derek Chang for their work on adding Trino support.

Lots of companies are trying to build text-to-SQL agents to help with business visibility, but most realize it's a hard problem to do well, especially with complex joins. Waii?+ LangGraph make it possible. ?? Read the blog and get the starter code, so you can build your analytics agent too! ?? https://lnkd.in/gE8Kr9KD