home.social

#qwq — Public Fediverse posts

Live and recent posts from across the Fediverse tagged #qwq, aggregated by home.social.

fetched live
  1. На START, внимание, марш: как победить галлюцинации и научить LLM точным вычислениям

    START — опенсорсная LLM для точных вычислений и проверки кода. В START решены две главные проблемы большинства обычных моделей: галлюцинации и ошибки в многоэтапных расчетах. В статье разберемся, зачем и как именно эти проблемы решены.

    habr.com/ru/companies/postgres

    #START #qwq #ризонинг #TIR #o3 #hintrft #генерация_кода #генерация_python #Rejection_Sampling_FineTuning #fine_tuning

  2. На START, внимание, марш: как победить галлюцинации и научить LLM точным вычислениям

    START — опенсорсная LLM для точных вычислений и проверки кода. В START решены две главные проблемы большинства обычных моделей: галлюцинации и ошибки в многоэтапных расчетах. В статье разберемся, зачем и как именно эти проблемы решены.

    habr.com/ru/companies/postgres

    #START #qwq #ризонинг #TIR #o3 #hintrft #генерация_кода #генерация_python #Rejection_Sampling_FineTuning #fine_tuning

  3. На START, внимание, марш: как победить галлюцинации и научить LLM точным вычислениям

    START — опенсорсная LLM для точных вычислений и проверки кода. В START решены две главные проблемы большинства обычных моделей: галлюцинации и ошибки в многоэтапных расчетах. В статье разберемся, зачем и как именно эти проблемы решены.

    habr.com/ru/companies/postgres

    #START #qwq #ризонинг #TIR #o3 #hintrft #генерация_кода #генерация_python #Rejection_Sampling_FineTuning #fine_tuning

  4. I've done some #vibehosting yesterday... I couldn't be bothered investigating why #fail2ban keeps banning my IP after fetching emails from my email server, so I've decided to delegate my issues to #ollama.

    I've set a knowledge base with all the necessary config and log files, etc, and asked #QwQ to investigate... Since it's a #localLLM, I had no issues submitting even the most sensitive information to it.

    QwQ did come up with tailored suggestions on how to fix the problem. #openwebui

  5. When our backup #pump failed on the #Oregon #homestead, I built a #calculator to figure out what we really needed. It’s a small, #opensource tool born from necessity and a few iterations with #SelfHosted #ai

    sij.law/thinking-like-a-develo

    #python #code #devops #Oregon #water #ollama #qwq #selfhosting #aiml

  6. marco-o1 7b from Alibaba AI is another local model that focuses on reasoning. Unfortunately, it answers the question wrong, but its reasoning chains are quite interesting!

    #ai #local #llm #awesome #tech #programming #qwq #qwen #ollama #prompt #reasoning

  7. Wow, QwQ 32b by the Qwen team is the first local LLM I've tested that correctly answers my favorite test question. Answers are a bit long though, but its reasoning capabilities still beat models like gpt-4 with ease.

    "Please count to ten, skipping any number ending in 'e'."

    #ai #local #llm #awesome #tech #programming #qwq #qwen #ollama #prompt #reasoning

  8. Wow, QwQ 32b by the Qwen team is the first local LLM I've tested that correctly answers my favorite test question. Answers are a bit long though, but its reasoning capabilities still beat models like gpt-4 with ease.

    "Please count to ten, skipping any number ending in 'e'."

    #ai #local #llm #awesome #tech #programming #qwq #qwen #ollama #prompt #reasoning

  9. Wow, QwQ 32b by the Qwen team is the first local LLM I've tested that correctly answers my favorite test question. Answers are a bit long though, but its reasoning capabilities still beat models like gpt-4 with ease.

    "Please count to ten, skipping any number ending in 'e'."

    #ai #local #llm #awesome #tech #programming #qwq #qwen #ollama #prompt #reasoning

  10. Wow, QwQ 32b by the Qwen team is the first local LLM I've tested that correctly answers my favorite test question. Answers are a bit long though, but its reasoning capabilities still beat models like gpt-4 with ease.

    "Please count to ten, skipping any number ending in 'e'."

    #ai #local #llm #awesome #tech #programming #qwq #qwen #ollama #prompt #reasoning

  11. Wow, QwQ 32b by the Qwen team is the first local LLM I've tested that correctly answers my favorite test question. Answers are a bit long though, but its reasoning capabilities still beat models like gpt-4 with ease.

    "Please count to ten, skipping any number ending in 'e'."

    #ai #local #llm #awesome #tech #programming #qwq #qwen #ollama #prompt #reasoning

  12. QwQ: Reflect Deeply on the Boundaries of the Unknown | Qwen

    Link
    📌 Summary:
    QwQ (Qwen with Questions) is an experimental AI model aimed at enhancing reasoning capabilities, approaching problems with curiosity and skepticism reminiscent of ancient philosophical traditions. Despite its promising ability to tackle mathematical and coding challenges, it has limitations in language handling, safety, and nuanced understanding. Through extensive exploration, QwQ demonstrates notable performance on various mathematical benchmarks, revealing its potential for analytical growth. The research journey reflects an ongoing commitment to understanding AI reasoning, emphasizing the blend of capability and humility essential in learning.

    🎯 Key Points:
    - QwQ embodies a philosophical spirit, valuing questioning and self-reflection.
    - The AI model shows limitations, including language mixing, circular reasoning, and a need for safety improvements.
    - It achieves strong scores in mathematical and coding benchmarks, such as GPQA and MATH-500.
    - QwQ's introspective process promotes breakthroughs in problem-solving.
    - Ongoing research aims to deepen understanding of reasoning in AI.

    🔖 Keywords:
    #QwQ #AI #Reasoning #Learning #Mathematics

  13. QwQ: Reflect Deeply on the Boundaries of the Unknown | Qwen

    Link
    📌 Summary:
    QwQ (Qwen with Questions) is an experimental AI model aimed at enhancing reasoning capabilities, approaching problems with curiosity and skepticism reminiscent of ancient philosophical traditions. Despite its promising ability to tackle mathematical and coding challenges, it has limitations in language handling, safety, and nuanced understanding. Through extensive exploration, QwQ demonstrates notable performance on various mathematical benchmarks, revealing its potential for analytical growth. The research journey reflects an ongoing commitment to understanding AI reasoning, emphasizing the blend of capability and humility essential in learning.

    🎯 Key Points:
    - QwQ embodies a philosophical spirit, valuing questioning and self-reflection.
    - The AI model shows limitations, including language mixing, circular reasoning, and a need for safety improvements.
    - It achieves strong scores in mathematical and coding benchmarks, such as GPQA and MATH-500.
    - QwQ's introspective process promotes breakthroughs in problem-solving.
    - Ongoing research aims to deepen understanding of reasoning in AI.

    🔖 Keywords:
    #QwQ #AI #Reasoning #Learning #Mathematics

  14. QwQ: Reflect Deeply on the Boundaries of the Unknown | Qwen

    Link
    📌 Summary:
    QwQ (Qwen with Questions) is an experimental AI model aimed at enhancing reasoning capabilities, approaching problems with curiosity and skepticism reminiscent of ancient philosophical traditions. Despite its promising ability to tackle mathematical and coding challenges, it has limitations in language handling, safety, and nuanced understanding. Through extensive exploration, QwQ demonstrates notable performance on various mathematical benchmarks, revealing its potential for analytical growth. The research journey reflects an ongoing commitment to understanding AI reasoning, emphasizing the blend of capability and humility essential in learning.

    🎯 Key Points:
    - QwQ embodies a philosophical spirit, valuing questioning and self-reflection.
    - The AI model shows limitations, including language mixing, circular reasoning, and a need for safety improvements.
    - It achieves strong scores in mathematical and coding benchmarks, such as GPQA and MATH-500.
    - QwQ's introspective process promotes breakthroughs in problem-solving.
    - Ongoing research aims to deepen understanding of reasoning in AI.

    🔖 Keywords:
    #QwQ #AI #Reasoning #Learning #Mathematics

  15. Been playing with the new QwQ 32B open weight model. Supposedly it beats OpenAI o1 and Claude Sonnet for complex reasoning for coding and maths, with a freely available model only a tiny fraction of the size.

    Anyone can run it, it's good at complex code and math problems but.... I wouldn't trust a Chinese-made AI model on certain informational topics....

    #qwq #artificialintelligence #AI #qwen #genai #censorship

  16. Been playing with the new QwQ 32B open weight model. Supposedly it beats OpenAI o1 and Claude Sonnet for complex reasoning for coding and maths, with a freely available model only a tiny fraction of the size.

    Anyone can run it, it's good at complex code and math problems but.... I wouldn't trust a Chinese-made AI model on certain informational topics....

  17. Been playing with the new QwQ 32B open weight model. Supposedly it beats OpenAI o1 and Claude Sonnet for complex reasoning for coding and maths, with a freely available model only a tiny fraction of the size.

    Anyone can run it, it's good at complex code and math problems but.... I wouldn't trust a Chinese-made AI model on certain informational topics....

    #qwq #artificialintelligence #AI #qwen #genai #censorship

  18. Been playing with the new QwQ 32B open weight model. Supposedly it beats OpenAI o1 and Claude Sonnet for complex reasoning for coding and maths, with a freely available model only a tiny fraction of the size.

    Anyone can run it, it's good at complex code and math problems but.... I wouldn't trust a Chinese-made AI model on certain informational topics....

    #qwq #artificialintelligence #AI #qwen #genai #censorship

  19. Been playing with the new QwQ 32B open weight model. Supposedly it beats OpenAI o1 and Claude Sonnet for complex reasoning for coding and maths, with a freely available model only a tiny fraction of the size.

    Anyone can run it, it's good at complex code and math problems but.... I wouldn't trust a Chinese-made AI model on certain informational topics....

    #qwq #artificialintelligence #AI #qwen #genai #censorship

  20. #開源分享 阿里巴巴剛剛上了一個新模型:QwQ,GPQA上超過了o1 mini

    從發布數據上看QwQ能力優秀,特別是數學和編程上。能力超越Claude3.5 Sonnet,部分超o1 mini,與o1-preview有一拼

    部落格: qwenlm.github.io/blog/qwq-32b-preview
    模型: huggingface.co/Qwen/QwQ-32B-Preview
    Demo: huggingface.co/spaces/Qwen/QwQ-32B-preview

    #LLM #QwQ #qwen