Downloads · 30 days
2
3% of all-time downloads
cas-smo_v32
tope1129/cas-smo_v32
cas-smo_v32 is a machine learning model from tope1129. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
Fork of cas-smo-v31 / garuda-labs/gen-04f5171856b6, retuned from pool fingerprints (no cascade score / no Toto2 training).
All-time downloads
59
Public
Repo size
—
Likes
0
Public
Repo makeup
Click a slice to open those files.
.py62.5 KB · 90%
At a glance
- Access
- Public
- Created
- Aug 13, 2026
- Updated
- Aug 13, 2026
- SHA
- 1deedd48
Model card
From the Hugging Face model README
cas-smo_v1
Fork of cas-smo-v31 / garuda-labs/gen-04f5171856b6, retuned from
pool fingerprints (no cascade score / no Toto2 training).
Target pool: dataset/snapshots/2026-08-09-block-8805600 (1945 series:
transport 28% / web_cloudops 23% / energy 16% / nature 16%). Use that
directory as --pool-dir if you later want a heat number.
Why this prior should score lower
The parent mix almost never emitted the two shapes that dominate the pool:
| Pool mass | Shape | Parent | cas-smo_v1 |
|---|---|---|---|
| ~21% GBFS-like docks | sticky integer occupancy, hold p50 0.83, cap ~19, mostly ±1 | k24 weight 0 | k24 0.15, hold mixture 0.28/0.84/0.98 |
| ~15% misskey + TFL counts | hourly integers, lag1 0.18–0.49 | triangle-wave k20 | noisy overdispersed Poisson k20 + heavier k11/k17 |
| cadence | periods 24 / 288 / 60 / 7 | table peaked at 7 and 96 | table peaked at 24 and 288 |
Also: 73% of the pool is integer-valued, so post-hoc rint was raised
(0.85/0.70 → 0.92/0.82), mixup is off, prefetch is 4.
Code vs config
_k20— triangle occupancy → noisy integer daily demand_k24— GBFS hold mixture + stronger daily tide_k11/_k17/_k25— periods and overdispersion for hourly / 5min_qsv06/_qsv07— seasonal table and harmonic pairsconfig.jsonp03— mass ontok24/k15/k11/k17/k20/k25
Fingerprint proxy (no training)
64-series sample vs the 1945-series pool, quantile L1 on hold / integer / lag-1 / lag-24 / zeros. Lower is a closer prior.
| garuda parent | cas-smo_v1 | pool | |
|---|---|---|---|
| total L1 | 0.650 | 0.383 | — |
| mean integer frac | 0.51 | 0.70 | 0.73 |
| mean lag-1 | 0.89 | 0.83 | 0.78 |
This is a prior match, not a trained geomean. Confirm with:
cascade score ./cas-smo_v1 --pool-dir ./dataset/snapshots/2026-08-09-block-8805600 --device cpu
- Created
- Aug 13, 2026
- Updated
- Aug 13, 2026