Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Run this model inference with full control and performance in your environment.
Get help setting up a custom Dedicated Endpoints.
Talk with our engineer to get a quote for reserved GPU instances with discounts.
README
License: apache-2.0Verification (free-running = self-generated latents)
| criterion | result |
|---|---|
| multi-step, EACH step load-bearing | corrupt any step -> chance (worst 0.078 vs 0.930) |
| parallel | K=3 cells per step |
| parallelism necessary | light-cone proof |
| load-bearing | ablate step1->prompt = 0.090 (chance) |
organism = 0.930. Generalization: held-out (fresh instances) = 1.000/1.000 (no memorization); depth (more steps than trained) = +1=0.00, +2=0.00 — this depth over-specialized — it generalizes across instances but not to deeper chains.
Controls
| intervention on the free-running latents | answer acc |
|---|---|
| intact | 1.000 |
| shuffle (permute latent positions) | 0.106 |
| cross-patch (swap in another instance's latents) | 0.113 |
Shuffle and cross-patch both collapse to chance (0.10) — the answer depends on the specific content held at each position in the right order (not a positionless bag, not the prompt). This is the signature of genuinely load-bearing latents.
Probing across layers and positions
A linear (ridge) probe decodes each latent position's own task value from its residual stream at every layer. The per-position state is linearly readable, peaking at layer 36 (mean decodability 0.90 across positions; chance 0.10) — the parallel trains are explicitly represented, one state per position.
Model provider
cds-jb
Model tree
Base
Qwen/Qwen3-8B
Adapter
this model
Modalities
Input
Text
Output
Text
Pricing
Dedicated Endpoints
View detailsSupported Functionality
Model APIs
Dedicated Endpoints
Container
More information