Cmpnd releases dspy-wsp for weight space programming in DSPy
Cmpnd released a prototype optimizer called dspy-wsp, which allows frontier models to edit small neural network weights directly without gradients or training loops.
FTMQ SI, written by our newsroom0 views

MarkTechPost reported that Cmpnd released dspy-wsp, a prototype optimizer for weight-space programming built for the DSPy framework. Weight-space programming allows a large language model to inspect and edit the weights of a smaller model directly inside a sandbox environment. The process operates without gradients, traditional training loops, or fine-tuning datasets. [1]
According to MarkTechPost, the new tool packages the functionality as an optimizer named WeightSpace. This module plugs directly into any standard DSPy program and metric. MarkTechPost reported that the system executes continuous trial-and-error edits on small model parameters while evaluating each iteration against target metrics. [1]
Prompt engineering involves structuring natural language inputs to guide generative AI outputs, as Wikipedia notes for context. Weight-space programming moves beyond natural language prompts by giving the frontier model direct access to binary neural network parameters. Cmpnd applied this technique to edit weights directly using an advanced model. [1][4]
In testing, GPT-6-Astra edited model weights directly and improved the accuracy of d1-omni-600M, according to MarkTechPost. Cmpnd reported that between $5 and $10 of API calls to GPT-6-Astra produced performance gains on every decision model evaluated in their trials. [1]
In short
- Cmpnd released dspy-wsp as a prototype DSPy optimizer for weight-space programming.
- The weight-space programming technique uses no gradients, training loops, or fine-tuning datasets.
- GPT-6-Astra edited weights directly to improve the accuracy of d1-omni-600M in Cmpnd tests.
- Each optimization run costs approximately $5 to $10 in API calls to GPT-6-Astra.
Sources
Every paragraph above points to the numbered items it rests on. Read the originals here.
- [1]Cmpnd Releases dspy-wsp: Weight-Space Programming Comes to DSPyMarkTechPost, 3d ago (the report this story comes from)
Background
- [2]Apollo program on Wikipedia
- [3]Atari Calculator on Wikipedia
- [4]Prompt engineering on Wikipedia
- [5]Developer releases tool to remove Apple Intelligence from macOS 27 FTMQ SI, 43d ago
Our newsroom writes these reports with the help of software, from the 5 sources listed and nothing else, and checks them against those sources. Facts can still be wrong or move on; the originals are the record. Spotted a mistake? Write to daniel@monsterkong.com.
Related from FTMQ SI
Earlier reports of ours on the same people and subjects.
- OpenAI releases GPT-6.2 Astra and Sol modelsIndustry Labs, 4h ago
- Open Chinese AI models approach closed frontier capabilities in network hacking testsTop Stories, 11d ago
- New GTFormer architecture replaces Transformer attention to process 4 million tokensResearch Papers, 17d ago
- DeepSeek Open Sources Router Worker Coding Agent Prototype Called dsh superagentModels & Products, 16d ago
- Developer releases tool to remove Apple Intelligence from macOS 27Top Stories, 43d ago
More in Models & Products
- Microsoft unveils Copilot Workspaces and ROI tools at Ignite 20264h ago
- Meta deploys LLM test system to catch software bugs in production1d ago
- Alibaba releases Qwen4-Max model with native vision MoE architecture1d ago
- NiCE launches next generation Cognigy platform for agentic contact center automation2d ago
- AI scanning tool uncovers dozens of Linux kernel bugs2d ago
- Bland releases Tofu 1 speech model for phone calls2d ago
Get the day in one email
Reports like this one, the top news of the last 24 hours, every morning. Free, one email a day; readers can comment under every report.
By signing up you agree to our terms and privacy policy. Unsubscribe any time.

Comments
Loading
Join the conversation
Comments are open to readers of our daily email: the top news of the last 24 hours, every morning, free. Sign up and the comment box opens.
Already on the list? Enter the same address and we will send a sign-in link.
By signing up you agree to our terms and privacy policy. Unsubscribe any time.