GEPA: prompt optimization can exceed RL performance
They used Qwen3-8B (which was not trained for math, coding, agency, etc.) and show that GEPA performed better than RL rollouts
paper: arxiv.org/abs/2507.19457
github: github.com/gepa-ai/gepa
DSPy docs: dspy.ai/api/optimize...
They used Qwen3-8B (which was not trained for math, coding, agency, etc.) and show that GEPA performed better than RL rollouts
paper: arxiv.org/abs/2507.19457
github: github.com/gepa-ai/gepa
DSPy docs: dspy.ai/api/optimize...
1. GEPA Overview - DSPy
The framework for programming—rather than prompting—language models.
dspy.ai
October 22, 2025 at 11:55 AM
GEPA: prompt optimization can exceed RL performance
They used Qwen3-8B (which was not trained for math, coding, agency, etc.) and show that GEPA performed better than RL rollouts
paper: arxiv.org/abs/2507.19457
github: github.com/gepa-ai/gepa
DSPy docs: dspy.ai/api/optimize...
They used Qwen3-8B (which was not trained for math, coding, agency, etc.) and show that GEPA performed better than RL rollouts
paper: arxiv.org/abs/2507.19457
github: github.com/gepa-ai/gepa
DSPy docs: dspy.ai/api/optimize...
Recommendation for @richards7370.bsky.social's debut album challenge:
www.youtube.com/watch?v=dsPY...
Soft Cell - Non-Stop Erotic Cabaret
#5debutalbums8084
www.youtube.com/watch?v=dsPY...
Soft Cell - Non-Stop Erotic Cabaret
#5debutalbums8084
Bedsitter (Remastered 2023)
YouTube video by Soft Cell - Topic
www.youtube.com
October 10, 2025 at 5:08 PM
Recommendation for @richards7370.bsky.social's debut album challenge:
www.youtube.com/watch?v=dsPY...
Soft Cell - Non-Stop Erotic Cabaret
#5debutalbums8084
www.youtube.com/watch?v=dsPY...
Soft Cell - Non-Stop Erotic Cabaret
#5debutalbums8084
No, it's a good idea! We just need something better than DSPy to do the initial variation, I guess.
October 9, 2025 at 5:32 PM
No, it's a good idea! We just need something better than DSPy to do the initial variation, I guess.
did you hear what he said about DSPy?? get him!! 😅
No, it's a good idea! We just need something better than DSPy to do the initial variation, I guess.
October 9, 2025 at 5:39 PM
did you hear what he said about DSPy?? get him!! 😅
Thank you everyone for your suggestions! This is how I ended up plotting my LLM system output vs human annotation vs LLM-as-a-judge evals. Extra thanks to @libbyheeren.bsky.social for boosting my original question. #databs #dataviz
October 8, 2025 at 11:52 AM
Thank you everyone for your suggestions! This is how I ended up plotting my LLM system output vs human annotation vs LLM-as-a-judge evals. Extra thanks to @libbyheeren.bsky.social for boosting my original question. #databs #dataviz
i feel like DSPy is beginning to occupy the Haskell tier
everyone: "DSPy is great, all LLM programming should be done like this"
narrator: "No real LLM programming is done this way"
everyone: "DSPy is great, all LLM programming should be done like this"
narrator: "No real LLM programming is done this way"
October 6, 2025 at 2:12 PM
i feel like DSPy is beginning to occupy the Haskell tier
everyone: "DSPy is great, all LLM programming should be done like this"
narrator: "No real LLM programming is done this way"
everyone: "DSPy is great, all LLM programming should be done like this"
narrator: "No real LLM programming is done this way"
If you've been trying to figure out DSPy - the automatic prompt optimization system - this talk by @dbreunig.bsky.social is the clearest explanation I've seen yet, with a very useful real-world case study www.youtube.com/watch?v=I9Zt...
My notes here: simonwillison.net/2025/Oct/4/d...
My notes here: simonwillison.net/2025/Oct/4/d...
Let the LLM Write the Prompts: An Intro to DSPy in Compound AI Pipelines
YouTube video by Databricks
www.youtube.com
October 4, 2025 at 11:05 PM
If you've been trying to figure out DSPy - the automatic prompt optimization system - this talk by @dbreunig.bsky.social is the clearest explanation I've seen yet, with a very useful real-world case study www.youtube.com/watch?v=I9Zt...
My notes here: simonwillison.net/2025/Oct/4/d...
My notes here: simonwillison.net/2025/Oct/4/d...
Why, I *have* been trying to figure out DSPy, thanks Simon! I've been looking for a good explanation, this is very useful.
October 4, 2025 at 11:19 PM
Why, I *have* been trying to figure out DSPy, thanks Simon! I've been looking for a good explanation, this is very useful.
AGI is just around the corner!
I'm learning to use DSPy with GEPA (Genetic-Pareto) prompt optimization. In GEPA a larger "teacher" LLM adjusts the prompt for a smaller "student" LM to perform a specific task as well as possible. The teacher will try many different prompts and evaluate the […]
I'm learning to use DSPy with GEPA (Genetic-Pareto) prompt optimization. In GEPA a larger "teacher" LLM adjusts the prompt for a smaller "student" LM to perform a specific task as well as possible. The teacher will try many different prompts and evaluate the […]
Original post on sigmoid.social
sigmoid.social
September 30, 2025 at 6:56 AM
AGI is just around the corner!
I'm learning to use DSPy with GEPA (Genetic-Pareto) prompt optimization. In GEPA a larger "teacher" LLM adjusts the prompt for a smaller "student" LM to perform a specific task as well as possible. The teacher will try many different prompts and evaluate the […]
I'm learning to use DSPy with GEPA (Genetic-Pareto) prompt optimization. In GEPA a larger "teacher" LLM adjusts the prompt for a smaller "student" LM to perform a specific task as well as possible. The teacher will try many different prompts and evaluate the […]
DSPy folks love GEPA, so here's a GEPA paper for anyone who wants to learn more.
Given any AI system containing one or more LLM prompts, GEPA samples system-level trajectories (e.g., reasoning, tool calls, and tool outputs) and reflects on them in natural language to diagnose problems,
Given any AI system containing one or more LLM prompts, GEPA samples system-level trajectories (e.g., reasoning, tool calls, and tool outputs) and reflects on them in natural language to diagnose problems,
September 28, 2025 at 12:28 AM
DSPy folks love GEPA, so here's a GEPA paper for anyone who wants to learn more.
Given any AI system containing one or more LLM prompts, GEPA samples system-level trajectories (e.g., reasoning, tool calls, and tool outputs) and reflects on them in natural language to diagnose problems,
Given any AI system containing one or more LLM prompts, GEPA samples system-level trajectories (e.g., reasoning, tool calls, and tool outputs) and reflects on them in natural language to diagnose problems,
📝 Summary:
The document provides information about "autoflow," an open-source GraphRAG (Knowledge Graph) built on TiDB Vector, LlamaIndex, and DSPy. It includes features such as conversational search, an embeddable JavaScript snippet, deployment options, and the tech stack used. The document (1/2)
The document provides information about "autoflow," an open-source GraphRAG (Knowledge Graph) built on TiDB Vector, LlamaIndex, and DSPy. It includes features such as conversational search, an embeddable JavaScript snippet, deployment options, and the tech stack used. The document (1/2)
November 24, 2024 at 4:01 PM
📝 Summary:
The document provides information about "autoflow," an open-source GraphRAG (Knowledge Graph) built on TiDB Vector, LlamaIndex, and DSPy. It includes features such as conversational search, an embeddable JavaScript snippet, deployment options, and the tech stack used. The document (1/2)
The document provides information about "autoflow," an open-source GraphRAG (Knowledge Graph) built on TiDB Vector, LlamaIndex, and DSPy. It includes features such as conversational search, an embeddable JavaScript snippet, deployment options, and the tech stack used. The document (1/2)
📰 DSPy – Programming–not prompting–LMs
💬 Users find DSPy somewhat rigid for practical tasks, especially with prior instructions—struggles with optimization persist. 🤔
https://news.ycombinator.com/item?id=42343692
💬 Users find DSPy somewhat rigid for practical tasks, especially with prior instructions—struggles with optimization persist. 🤔
https://news.ycombinator.com/item?id=42343692
December 6, 2024 at 9:00 PM
📰 DSPy – Programming–not prompting–LMs
💬 Users find DSPy somewhat rigid for practical tasks, especially with prior instructions—struggles with optimization persist. 🤔
https://news.ycombinator.com/item?id=42343692
💬 Users find DSPy somewhat rigid for practical tasks, especially with prior instructions—struggles with optimization persist. 🤔
https://news.ycombinator.com/item?id=42343692
Converting AI programs to DSPy is always beneficial though!
August 31, 2025 at 1:57 AM
Converting AI programs to DSPy is always beneficial though!
Систематическая инженерия подсказок LLM с использованием оптимизации DSPy
Статья представляет собой путешествие в захватывающую и быстро развивающуюся науку итерации запросов языковой модели крупного масштаба (LLM), которая является фундаментальной частью операций с яз…
#ai #llm #promptengineering
Статья представляет собой путешествие в захватывающую и быстро развивающуюся науку итерации запросов языковой модели крупного масштаба (LLM), которая является фундаментальной частью операций с яз…
#ai #llm #promptengineering
Systematic LLM Prompt Engineering Using DSPy Optimization
towardsdatascience.com
August 26, 2025 at 12:04 PM
Систематическая инженерия подсказок LLM с использованием оптимизации DSPy
Статья представляет собой путешествие в захватывающую и быстро развивающуюся науку итерации запросов языковой модели крупного масштаба (LLM), которая является фундаментальной частью операций с яз…
#ai #llm #promptengineering
Статья представляет собой путешествие в захватывающую и быстро развивающуюся науку итерации запросов языковой модели крупного масштаба (LLM), которая является фундаментальной частью операций с яз…
#ai #llm #promptengineering
DSPyのサンプルを簡単に試してみたけれども高レベルなLangChainを使っているような気分で簡単な処理を手軽に生成AIにやってもらう目的だったりテストでサクッとアイデアを試したりする用途でもなかなか便利に使えそう
October 18, 2025 at 9:20 AM
DSPyのサンプルを簡単に試してみたけれども高レベルなLangChainを使っているような気分で簡単な処理を手軽に生成AIにやってもらう目的だったりテストでサクッとアイデアを試したりする用途でもなかなか便利に使えそう
DSPyは、プロンプトを直接いじる代わりに宣言的なPythonコードでLLMアプリを構築し、自動でプロンプトや重みを最適化してくれるフレームワークらしい。複雑なRAGパイプラインやエージェント開発に役立ちそう。
yug1224 starred stanfordnlp/dspy
https://github.com/stanfordnlp/dspy
yug1224 starred stanfordnlp/dspy
https://github.com/stanfordnlp/dspy
GitHub - stanfordnlp/dspy: DSPy: The framework for programming—not prompting—language models
github.com
October 26, 2025 at 6:24 AM
DSPyは、プロンプトを直接いじる代わりに宣言的なPythonコードでLLMアプリを構築し、自動でプロンプトや重みを最適化してくれるフレームワークらしい。複雑なRAGパイプラインやエージェント開発に役立ちそう。
yug1224 starred stanfordnlp/dspy
https://github.com/stanfordnlp/dspy
yug1224 starred stanfordnlp/dspy
https://github.com/stanfordnlp/dspy
Great point, perhaps an automated prompt optimisation approach (e.g. dspy) with a small dataset of examples for priming/conditioning the model would work well. That said, these search approaches can be very expensive!
December 7, 2024 at 6:52 PM
Great point, perhaps an automated prompt optimisation approach (e.g. dspy) with a small dataset of examples for priming/conditioning the model would work well. That said, these search approaches can be very expensive!
Whether you’re tuning a prompt or training a model, MLflow helps you understand what’s happening under the hood — so you can iterate faster and build better. ✅
MLflow documentation ➡️ mlflow.org/docs/latest/...
#mlflow #dspy #opensource #oss
MLflow documentation ➡️ mlflow.org/docs/latest/...
#mlflow #dspy #opensource #oss
April 21, 2025 at 7:22 PM
Whether you’re tuning a prompt or training a model, MLflow helps you understand what’s happening under the hood — so you can iterate faster and build better. ✅
MLflow documentation ➡️ mlflow.org/docs/latest/...
#mlflow #dspy #opensource #oss
MLflow documentation ➡️ mlflow.org/docs/latest/...
#mlflow #dspy #opensource #oss
www.youtube.com/watch?v=dsPY... Today in WJSN History they performed "Last Sequence" on Inkigayo #WJSN #WJSNHistory #LastSequence
WJSN(우주소녀) - Last Sequence @인기가요 inkigayo 20220710
YouTube video by SBSKPOP X INKIGAYO
www.youtube.com
July 10, 2025 at 8:19 PM
www.youtube.com/watch?v=dsPY... Today in WJSN History they performed "Last Sequence" on Inkigayo #WJSN #WJSNHistory #LastSequence
Fighting with postgres and langgraph this morning. Getting to the point where I just want to rip langgraph out and build my agent out of DSPy because what I'm dealing with is fucking insane.
August 18, 2025 at 3:10 PM
Fighting with postgres and langgraph this morning. Getting to the point where I just want to rip langgraph out and build my agent out of DSPy because what I'm dealing with is fucking insane.
DSPyがとても気になるけどAmazon Bedrockとかでも使えるといいな
dspy.ai
dspy.ai
DSPy
The framework for programming—rather than prompting—language models.
dspy.ai
October 18, 2025 at 3:54 AM
DSPyがとても気になるけどAmazon Bedrockとかでも使えるといいな
dspy.ai
dspy.ai
DSPyはプロンプトエンジニアリングを減らす可能性を持つと話されている。完全に不要になるわけではないが、使い心地がよく学びやすいと評価されている。
開発者はPFNのChainerに触れた経験とPyTorchの進化を思い出しながら、学びをまとめたブログを公開した。 zenn.dev #news
開発者はPFNのChainerに触れた経験とPyTorchの進化を思い出しながら、学びをまとめたブログを公開した。 zenn.dev #news
プロンプトエンジニアリングを終わらせるDSPy
zenn.dev
October 9, 2025 at 9:06 AM
Looks like it's primarily in the API. Synalinks is based on Keras, where dspy was based on Pytorch
synalinks.github.io/synalinks/FA...
synalinks.github.io/synalinks/FA...
FAQ - Synalinks
synalinks.github.io
March 4, 2025 at 3:21 AM
Looks like it's primarily in the API. Synalinks is based on Keras, where dspy was based on Pytorch
synalinks.github.io/synalinks/FA...
synalinks.github.io/synalinks/FA...