deepakdsoni
/

antahkarana-v1

@@ -51,9 +51,9 @@ re-evaluated), not training-time logs. 11 models, 7 capabilities each, run on a
 of 10 and 20 tasks. The agent both *remembers* (forgetting → ~0.01–0.05) and *learns better* (accuracy
 roughly **doubles** vs the naive baseline, which collapses to its last task).
-![forgetting](plots/bench_forgetting.png)
-![scorecard](plots/bench_scorecard.png)
 ## 4. Per-model detail (live inference)
@@ -88,7 +88,7 @@ twice as many tasks to retain with more interference. Reduction drops to **12.5
 **lower than the 10-task runs, and honestly so** — yet all 20/20 tasks remain above chance and the model still
 abstains/calibrates cleanly. A >12× cut over a 20-task lifelong run is a strong result.
-![retention](plots/bench_retention.png)
 *Per-task accuracy after the full stream — every task stays well above chance; the agent does not collapse
 to its last task (which is what the naive baseline does).*
@@ -97,7 +97,7 @@ to its last task (which is what the naive baseline does).*
 more it abstains — a desirable, safety-aligned property. Coverage is correspondingly lower on hard sets
 (a deliberate trade: it commits less, but is right when it does).
-![pramana](plots/bench_pramana.png)
 **Calibration.** Temperature scaling reduces ECE everywhere (e.g. 0.172 → 0.023), so reported confidence can
 be trusted — the precondition for the abstention gate to be meaningful.
@@ -107,7 +107,7 @@ life** (e.g. 0.92 → 0.37; 0.77 → 0.14 on the 20-task run), the agent **ages
 (brahmacarya → gṛhastha → vānaprastha → saṃnyāsa), and **turīya drift stays bounded** (~0.50–0.59). You can
 read the agent's regulation as it lives — not a black box.
-![mindstate](plots/bench_mindstate.png)
 ## 6. The one 6/7, explained (no goalpost-moving)

 of 10 and 20 tasks. The agent both *remembers* (forgetting → ~0.01–0.05) and *learns better* (accuracy
 roughly **doubles** vs the naive baseline, which collapses to its last task).
+![forgetting](https://huggingface.co/deepakdsoni/antahkarana/resolve/main/plots/bench_forgetting.png)
+![scorecard](https://huggingface.co/deepakdsoni/antahkarana/resolve/main/plots/bench_scorecard.png)
 ## 4. Per-model detail (live inference)
 **lower than the 10-task runs, and honestly so** — yet all 20/20 tasks remain above chance and the model still
 abstains/calibrates cleanly. A >12× cut over a 20-task lifelong run is a strong result.
+![retention](https://huggingface.co/deepakdsoni/antahkarana/resolve/main/plots/bench_retention.png)
 *Per-task accuracy after the full stream — every task stays well above chance; the agent does not collapse
 to its last task (which is what the naive baseline does).*
 more it abstains — a desirable, safety-aligned property. Coverage is correspondingly lower on hard sets
 (a deliberate trade: it commits less, but is right when it does).
+![pramana](https://huggingface.co/deepakdsoni/antahkarana/resolve/main/plots/bench_pramana.png)
 **Calibration.** Temperature scaling reduces ECE everywhere (e.g. 0.172 → 0.023), so reported confidence can
 be trusted — the precondition for the abstention gate to be meaningful.
 (brahmacarya → gṛhastha → vānaprastha → saṃnyāsa), and **turīya drift stays bounded** (~0.50–0.59). You can
 read the agent's regulation as it lives — not a black box.
+![mindstate](https://huggingface.co/deepakdsoni/antahkarana/resolve/main/plots/bench_mindstate.png)
 ## 6. The one 6/7, explained (no goalpost-moving)