Cmpnd Releases dspy-wsp Letting Frontier LLMs Patch Small Model Weights

1 reports archived1 independent sourcesupdated 11/15/2026, 00:53:52

Synthesis & Latest Updates

1 Sources Cross-Validated

Cmpnd open-sourced dspy-wsp, a DSPy-based weight-space programming optimizer in which a frontier writer LLM inspects a student model's tensors in a sandbox, forms hypotheses and patches a few weights, with no gradients, training loop or fine-tuning data. On the Typed Decisions benchmark, d1-omni-600M rose from 50.9% to 66.9% with WeightSpace alone and 69.1% when stacked with ReAnchor, though a 9B model gained only 1.9 points and each run cost $5.42 to $7.63 over 78 to 167 minutes.

LATEST/Cmpnd open-sourced dspy-wsp, a DSPy-based weight-space programming optimizer in which a frontier writer LLM inspects a student model's tensors in a sandbox, forms hypotheses and patches a few weights, with no gradients, training loop or fine-tuning data. On the Typed Decisions benchmark, d1-omni-600M rose from 50.9% to 66.9% with WeightSpace alone and 69.1% when stacked with ReAnchor, though a 9B model gained only 1.9 points and each run cost $5.42 to $7.63 over 78 to 167 minutes.

Hourly heat curve

7 fully observed hours

Now

15.8

Peak

17.901:00

24h change

–

The line compares only sources observed throughout; gap hours are interpolated to keep the trend continuous.

Coverage timeline

Total 1 reports · Latest first

Marktechpost (@Marktechpost)T2Nov 15, 00:53·74 pts
Cmpnd Releases dspy-wsp Letting Frontier LLMs Patch Small Model Weights
- A frontier LLM can read and patch small-model tensors in a sandbox, bypassing gradients and fine-tuning data
- d1-omni-600M gained 16.1 points on Typed Decisions, reaching 18.3 when stacked with ReAnchor
- Gains shrink with scale, with a 9B model gaining only 1.9 points at roughly $5 to $8 per run
View Related Reports & Analysis Original Source