{"runtimes":[{"vocab":8192,"species":["soma_rma_v1"],"tokenizer":"soma-8k-v1","checkpoint_format":"soma-rma-v1-pytorch","id":"rma-v1","runtime_id":"soma.rma_v1","family":"soma","name":"rma v1","headline":"residual metabolic aperture","one_liner":"the residual metabolic aperture: an 8k-token alphabet read through exact phi-spaced temporal traces and nested cumulative apertures, with io2 choosing its own prediction stride and plasticity from the spectrum of its residual. bounded state, fixed matrix energy, no optimizer moments. trains locally in the soma harness or in the cloud through logOS.","status":"latest","visibility":"public","runtime_version":"1.8.1","released":"2026-10-04","release_count":5,"downloads":{"spec_page":"/runtimes/rma-v1","zip":"/api/runtimes/rma-v1/download/zip","spec":"/api/runtimes/rma-v1/download/spec"},"files":[{"name":"soma-rma-v1-1.8.1.tar.gz","kind":"bundle","platform":"linux","media_type":"application/octet-stream","bytes":161584,"sha256":"ddbe58fdecad98790c75955cbb3c5d1e93dbdb99906ace04f78f1bc5289b7ee0","verified":true,"download_count":4,"url":"/api/runtimes/rma-v1/releases/1.8.1/files/soma-rma-v1-1.8.1.tar.gz"}]},{"vocab":1024,"species":["soma_v14_adaptive_trace_branch_mixer"],"token_dim":384,"tokenizer":"soma-1k-v1","quickstart":{"unix":"unzip soma_v14.zip\ncd soma-v14-script-bundle\n./soma                # mac / linux — bootstraps a venv on first run","manual":"pip install -r requirements.txt\npython soma_v14.py"},"default_shape":{"lr":"auto 3e-4","bands":48,"capacity":"medium","model_dim":1536,"batch_size":512,"parameters":65508870,"stride_cap":256,"trunk_depth":4,"expansion_dim":3072,"decimation_gain":1.0,"pulse_scheduling":"spectral (enabled)","lr_reference_batch":1024,"radial_compression":"soft, enabled"},"bundle_summary":"soma v14 as a python script bundle. adaptive trace-branch mixer — species soma_v14_adaptive_trace_branch_mixer, tokenizer soma-1k-v1 (vocab 1024), 48 exact temporal bands on an adaptive log-time lattice that io2 conservatively redistributes without shortening the horizon, three scale blocks (depthwise-5 conv → rank-64 low-rank symbol mixer → boundary-conditioned residual gates) → twelve fixed gaussian summaries → 1,536-wide context → four serial residual swiglu blocks → 384-dim branch read through the tied token dictionary with the empirical log-prior added. training is next-token cross entropy with adamw (auto lr 3e-4, reference batch 1024) inside canonical weighted-prediction quanta, io2 spectral pulse scheduling, soft radial compression, decimation gain, checkpoint contract with model_id / revision_id / parent_revision_id / root_lineage_id and a .revision.json publication guard. medium capacity default (65.5M params). the zip contains soma_v14.py, the checked-in soma-1k-v1 tokenizer, trace_branch_mixer/, soma_gui.py, soma_train_worker.py, soma_loop.py, streams/, corpus builders, cloud/, tests/, requirements.txt, soma_v14_spec.md, README.md, logOS_v14_web_handover.md and the on-device speech helpers (soma_voice.py + macos_soma_speech.swift + soma-speech binary). v13 trace-branch ancestors are accepted; older species are rejected. see /runtimes for every runtime we've ever shipped, including posthumous v12.3.","checkpoint_format":"soma-v14-pytorch","checkpoint_version":14,"id":"v14","runtime_id":"soma.v14","family":"soma","name":"soma v14","headline":"adaptive trace-branch mixer","one_liner":"the exact trace bank remains a nonlearned float64 float bank; a trace branch mixer organises its temporal field into a compact 1,536-dim present; four serial residual swiglu blocks compose. the temporal lens is now an adaptive lattice on log time, conservatively redistributed by io2.","status":"posthumous","visibility":"public","runtime_version":"14.0.0-dev.2","released":"2026-08-11","release_count":1,"downloads":{"spec_page":"/runtimes/v14","dmg":"/api/runtimes/v14/download/dmg","zip":"/api/runtimes/v14/download/zip","py":"/api/runtimes/v14/download/py","spec":"/api/runtimes/v14/download/spec"},"files":[{"name":"soma.dmg","kind":"app","platform":"macos","media_type":"application/x-apple-diskimage","bytes":154375370,"sha256":"29d660d1577a739a4ae4bbcfb982565a3d791c16316afac9a8495f52ef17e5e1","verified":true,"download_count":10,"url":"/api/runtimes/v14/releases/14.0.0-dev.2/files/soma.dmg"},{"name":"soma_v14.zip","kind":"bundle","platform":"any","media_type":"application/zip","bytes":278453,"sha256":"fb33c4980a53c6560f2eb0ae9453d53d40f5661d14d2f9b8b7cb67f1ac84f276","verified":true,"download_count":5,"url":"/api/runtimes/v14/releases/14.0.0-dev.2/files/soma_v14.zip"},{"name":"soma_v14.py","kind":"script","platform":"any","media_type":"text/x-python","bytes":57236,"sha256":"a7ffba701ce348bd5e15cbfa707acf210b3fff3573d1a6a3725337b54102fac5","verified":true,"download_count":5,"url":"/api/runtimes/v14/releases/14.0.0-dev.2/files/soma_v14.py"}]},{"vocab":8192,"species":["soma_v12_3_token_residual"],"token_dim":256,"tokenizer":"soma-8k-v1","default_shape":{"lr":"auto 2e-3","bands":16,"capacity":null,"model_dim":4096,"batch_size":512,"parameters":70300000,"stride_cap":256,"trunk_depth":4,"decimation_gain":1.0,"pulse_scheduling":null,"lr_reference_batch":null,"radial_compression":"radial plasticity (enabled)"},"checkpoint_format":"soma-v12.3-pytorch","checkpoint_version":null,"id":"v12.3","runtime_id":"soma.v12.3","family":"soma","name":"soma v12.3","headline":"token-native continual spectral learner","one_liner":"a fixed 8k sentencepiece dictionary feeds a token trace bank whose retained identity is projected through the current learned dictionary each batch; a serial budget-residual mlp composes over the projected present. radial plasticity compression, depth saturation, decimation gain.","status":"posthumous","visibility":"public","runtime_version":"12.3.17","released":"2026-08-01","release_count":1,"downloads":{"spec_page":"/runtimes/v12.3","dmg":"/api/runtimes/v12.3/download/dmg","zip":"/api/runtimes/v12.3/download/zip","py":"/api/runtimes/v12.3/download/py","spec":"/api/runtimes/v12.3/download/spec"},"files":[{"name":"soma.dmg","kind":"app","platform":"macos","media_type":"application/x-apple-diskimage","bytes":202371690,"sha256":"582bfcd81a7fd9d17234953d904ba2f5e6a7ea1d452a1f6e903d81937b4c2836","verified":true,"download_count":1,"url":"/api/runtimes/v12.3/releases/12.3.17/files/soma.dmg"},{"name":"soma_v12_3.zip","kind":"bundle","platform":"any","media_type":"application/zip","bytes":277989,"sha256":"23e6ef5c53ef717583a6a48abe589217ab328c9c1db0869bb4a11576f3e3b481","verified":true,"download_count":1,"url":"/api/runtimes/v12.3/releases/12.3.17/files/soma_v12_3.zip"},{"name":"soma_v12_3.py","kind":"script","platform":"any","media_type":"text/x-python","bytes":70665,"sha256":"f1b738e19e4be745ffc4f95c661c002ab0e4ee32e5d4b16d8eedc120d6824767","verified":true,"download_count":1,"url":"/api/runtimes/v12.3/releases/12.3.17/files/soma_v12_3.py"}]}]}