README: tidy two lines
Browse files
README.md
CHANGED
|
@@ -154,7 +154,7 @@ The easy tier and the first 48 standard rows ran on the charger.
|
|
| 154 |
matched tokens 50/50, answer slots 50/50 and option argmax 50/50 on both bundles, including the
|
| 155 |
40- and 255-option rows. int8lin max |Δp| was **0.0226** (`r05-dry`), mean **0.0009**, and max
|
| 156 |
|Δabstain| **0.0222**; fp16 was **0.0051** (`r05-task`), **0.0003**, and **0.0165** respectively.
|
| 157 |
-
These are kit measurements
|
| 158 |
table above. Median int8lin wall time per fixture question was **354 ms** over the 50 rows, two to
|
| 159 |
three questions per state (a question on a new state pays for the whole state); the 255-option row (1,449 tokens, S=1 prefill) took **7.2 s**. A
|
| 160 |
three-question Thai ticket took **351 / 316 / 429 ms** for its 57-, 63- and 83-token rows; the
|
|
@@ -175,7 +175,7 @@ argmax **50/50**, the 40- and 255-option rows included (the 1,449-token row fits
|
|
| 175 |
raw and **0.7249** mean family balanced accuracy — the Mac's figures exactly — at **2,074 ms** median per decision.
|
| 176 |
Cooled to thermal state "nominal" (2026-09-24, `decide-cli bench --repeat 3`, a 111-token state and eight questions):
|
| 177 |
**1,527 ms per decision** with the state shared, 1,713 from scratch, 13.5 s for the state and its eight (0 tokens
|
| 178 |
-
reused). Load 6.6 s.
|
| 179 |
[`gate-openthai-systemone-iphone-parity.json`](https://github.com/john-rocky/coreai-model-zoo/blob/main/models/openthai-systemone/gate-openthai-systemone-iphone-parity.json).
|
| 180 |
|
| 181 |
## Bundle
|
|
|
|
| 154 |
matched tokens 50/50, answer slots 50/50 and option argmax 50/50 on both bundles, including the
|
| 155 |
40- and 255-option rows. int8lin max |Δp| was **0.0226** (`r05-dry`), mean **0.0009**, and max
|
| 156 |
|Δabstain| **0.0222**; fp16 was **0.0051** (`r05-task`), **0.0003**, and **0.0165** respectively.
|
| 157 |
+
These are kit measurements, separate from the Python-runtime
|
| 158 |
table above. Median int8lin wall time per fixture question was **354 ms** over the 50 rows, two to
|
| 159 |
three questions per state (a question on a new state pays for the whole state); the 255-option row (1,449 tokens, S=1 prefill) took **7.2 s**. A
|
| 160 |
three-question Thai ticket took **351 / 316 / 429 ms** for its 57-, 63- and 83-token rows; the
|
|
|
|
| 175 |
raw and **0.7249** mean family balanced accuracy — the Mac's figures exactly — at **2,074 ms** median per decision.
|
| 176 |
Cooled to thermal state "nominal" (2026-09-24, `decide-cli bench --repeat 3`, a 111-token state and eight questions):
|
| 177 |
**1,527 ms per decision** with the state shared, 1,713 from scratch, 13.5 s for the state and its eight (0 tokens
|
| 178 |
+
reused). Load 6.6 s. Record:
|
| 179 |
[`gate-openthai-systemone-iphone-parity.json`](https://github.com/john-rocky/coreai-model-zoo/blob/main/models/openthai-systemone/gate-openthai-systemone-iphone-parity.json).
|
| 180 |
|
| 181 |
## Bundle
|