.txt Engineering

Most-linked posts?

1. Structured generation for LLMs performs better - a detailed rebuttal
   3 people · 5 mentions · first linked 2024-11-24
2. Coalescence: Making LLM inference 5x faster
   3 people · 3 mentions · first linked 2024-02-02
3. How fast can CFG-Structured generation be?
   2 people · 2 mentions · first linked 2024-04-12
4. Coding for Structured Generation with LLMs
   2 people · 2 mentions · first linked 2024-06-26
5. Beating GPT-4 with open source
   2 people · 2 mentions · first linked 2024-06-18

What Hacker News says?

1. Der_Einzige · 2025-01-01

   You should read the follow-up work from Entropix folks, or reflect on the extremely high review scores min_p is getting, or look at the fact the even trivial shit like top_k=2 + temperature = max_int works as evidence that models do in fact "have access to the semantics of what they're sampling" via the ordering of their logprobs. DRY does in fact solve repetition issues. You're not using the right settings with it.…

   Comment linking /say-what-you-mean.html

   on “Things we learned about LLMs in 2024”
2. Homunculiheaded · 2024-03-15

   Submitted Structured Generation Improves LLM Performance: GSM8K Benchmark as a story · 11 points

   HN discussion
3. Der_Einzige · 2025-01-30

   For anyone who thinks it isn't "solved", outlines debunked the paper which claims that "structured generation harms creativity": https://blog.dottxt.co/say-what-you-mean.html

   Comment linking /say-what-you-mean.html

   on “Mistral Small 3”
4. Homunculiheaded · 2024-04-12

   Submitted How fast can CFG-Structured generation be? as a story · 7 points

   HN discussion
5. crystal_revenge · 2024-12-07

   There was a paper going around claiming that structured outputs did hurt the quality of the output, but it turns out their experiment setup was laughably bad [0]. It looks like, so long as you're reasonable with the prompting, you tend to get better outputs when using structure. 0. https://blog.dottxt.co/say-what-you-mean.html

   Comment linking /say-what-you-mean.html

   on “Structured Outputs with Ollama”
6. Homunculiheaded · 2024-02-02

   Submitted Coalescence: Making LLM inference 5x faster as a story · 6 points

   HN discussion
7. remilouf · 2024-04-05

   Looks like it’s quite the opposite: http://blog.dottxt.co/performance-gsm8k.html

   Comment linking /performance-gsm8k.html

   on “Tool Use (function calling)”
8. benchmarkist · 2024-11-25

   Submitted Beating GPT-4 with open source as a story · 5 points

   HN discussion
9. Der_Einzige · 2025-01-14

   BTW - Structured/Constrained Generation is the KEY to making AI agents better/scary good. Without it, you're leaving so much on the table. This library is awesome for augmenting that capability!!!! Also, if you're "studying LLM based chess" and you don't use dynamic grammar's to enforce that models can only make "valid" moves at each time step, you're research is basically invalid. And don't meme me with claims that…

   Comment linking /say-what-you-mean.html

   on “Show HN: Value likelihoods for OpenAI structured output”
10. Homunculiheaded · 2024-06-18

    Submitted Beating GPT-4 with Open Source as a story · 5 points

    HN discussion
11. yeesian · 2024-03-07

    Submitted Coalescence: Making LLM inference 5x faster as a story · 5 points

    HN discussion
12. Homunculiheaded · 2024-07-19

    Submitted Making Structured Generation Faster Than Unstructured as a story · 4 points

    HN discussion
13. Homunculiheaded · 2024-06-26

    Submitted Coding for Structured Generation with LLMs as a story · 4 points

    HN discussion
14. sebg · 2024-06-27

    Submitted Coding for Structured Generation with LLMs as a story · 3 points

    HN discussion
15. remilouf · 2025-07-10

    Submitted Every AI Integration Is Held Together with Parsing Logic and Prayer as a story · 2 points

    HN discussion
16. sebg · 2025-03-04

    Submitted Building an LLM-Powered Web App to Generate Gift Ideas as a story · 2 points

    HN discussion
17. bfelbo · 2024-11-24

    Submitted Structured generation for LLMs performs better - a detailed rebuttal as a story · 2 points

    HN discussion
18. Homunculiheaded · 2024-05-07

    Submitted Prompt Efficiency – Getting 8-shot LLM performance from 1-shot as a story · 2 points

    HN discussion
19. btwillard · 2024-04-14

    Submitted How fast can grammar-structured LLM generation be? as a story · 2 points

    HN discussion
20. remilouf · 2024-02-04

    Submitted Tokenization Is a Problem for LLMs as a story · 2 points

    HN discussion
21. PaulHoule · 2024-12-04

    Submitted Coalescence: Making LLM inference 5x faster as a story · 1 points

    HN discussion
22. cpfiffer · 2024-11-07

    Submitted Extracting Financial Data from SEC 10-K Filings with LLMs as a story · 1 points

    HN discussion

Search all of HN for blog.dottxt.co