Date: 2026-03-08 Model: google/gemma-3-1b-it (262,144 vocab) Eval dataset: wikitext-2-raw-v1 validation (254,828 positions) Code: voltropy/shortlist@8168cac
We discovered that a static, frequency-ranked token set with a simple margin-based fallback to full-vocab scoring achieves better parity than a trained neural router, with zero parameters, zero training, and zero inference-time routing.
Date: 2026-03-02
Machine: A1 (216.81.248.152), NVIDIA A100-SXM4-80GB, PyTorch 2.10.0+cu126, CUDA 12.6
Code: monarch repo, commit aa3bb6f (token-local swap-FFN state)
Checkpoints: Trained baseline and hybrid1 checkpoints from s3://voltcode-artifacts-17f9c348/runs/monarch-swap-ffn/20260302/
Results: s3://voltcode-artifacts-17f9c348/runs/swap-ffn-bench/a1-compiled-20260302/
Detailed execution trace comparing the fused Triton kernel path vs the unfused PyTorch/cuBLAS path for Swap-FFN decode inference.
Config: d_model=1024, d_ff=4096, core_count=512, n_events=8, k_ffn=2, d_router=64, collapse_k=8, collapse_r=224
Hardware: A100-SXM4-80GB, 108 SMs, 40 MB L2 cache
| console.log("Verification test"); |
| console.log("Hello from MCP server!"); |
| |pass &noun | |
| :- %k %lard %base !< shed:khan | |
| %+ slap (slop .^(vase %ca %/lib/strandio/hoon) !>(..zuse)) | |
| !, *hoon | |
| =* strandio - | |
| =, strand=strand:rand | |
| ^- shed:khan | |
| =/ m (strand ,vase) | |
| |- ^- form:m | |
| ;< =bowl:rand bind:m get-bowl:strandio |
| => | |
| |% | |
| +$ ulam | |
| $~ [%coin *coin] | |
| :: leaves | |
| :: | |
| $% [%coin =coin] :: atom, noun, or compound coin | |
| [%path =pith] :: hoon path syntax | |
| [%page =mark noun=*] :: %foo|bar, bar is coin blob | |
| :: |
%json mark from 414 will not compile on 413, and this will not be caught at build time.
%json mark file in your app desk compiles on 413, and test your %json mark conversions to make sure they don't crash at runtime due to this change.+en-json, +de-json). apex:de-json:html, part of the JSON parser internals, isn't there any more.%gall scry namespace has been changed:
%gu, to check if an app is an running, has a new path format:
.^(? %gu /=dojo=).^(? %gu /=dojo=/$)
| /+ default-agent, verb, dbug | |
| :: | |
| |% | |
| +$ poke | |
| $% [%grow =case =spur =noun] | |
| [%tomb =case =spur] | |
| [%cull =case =spur] | |
| == | |
| -- | |
| :: |
This Kelvin requires more work to update than some previous Kelvins because the $foam type was deleted from Lull. This means if your app imports the full base-dev folder as a dependency the way most apps do and your app imports /sur/hood/hoon (as a lot of apps do), you will need to update that dependency for 414 by copying in the most recent files, and also drop support for older Kelvins by deleting their entries from /sys/kelvin and replacing them all with just [%zuse 414]. If your app does not have any files that import /sur/hood/hoon with Ford runes, you might be able to support 414 and 415 without modifying code -- if so, please test your app to be sure.
Note that Clay can still handle later commits targeted at 415, so as an app dev, you can maintain both 414 and 415 -- the constraint is that each commit to your app desk can only support one or the other if you are using the standard /sur/hood/hoon. Supporting both in the same stream of commits to