Instructions to use primce/tiny-aya-hinglish-lora with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- PEFT
How to use primce/tiny-aya-hinglish-lora with PEFT:
Task type is invalid.
- Notebooks
- Google Colab
- Kaggle
State the oracle call count per split
Browse files
README.md
CHANGED
|
@@ -56,7 +56,8 @@ seed adapters are kept for the ablation and variance controls.
|
|
| 56 |
round-2 adapter, **18 fixed, 2 broken, p = 0.0004**; against the round-1 adapter, 7 fixed,
|
| 57 |
3 broken, p = 0.34 (directional, not significant).
|
| 58 |
- Tool calls per task: **3.57** on hard-30, exactly the oracle's count (base: 7.03). On hard-90:
|
| 59 |
-
3.38 against a base 7.24
|
|
|
|
| 60 |
- Tool error rate **0.00%** on both splits, loop collapse **0.00%**, redundant calls **0.00%**,
|
| 61 |
parse failures **0.00%** (base: 22.3% tool errors, 16 of 30 episodes never finishing,
|
| 62 |
`overcall` 1.08).
|
|
|
|
| 56 |
round-2 adapter, **18 fixed, 2 broken, p = 0.0004**; against the round-1 adapter, 7 fixed,
|
| 57 |
3 broken, p = 0.34 (directional, not significant).
|
| 58 |
- Tool calls per task: **3.57** on hard-30, exactly the oracle's count (base: 7.03). On hard-90:
|
| 59 |
+
3.38 against an oracle 3.41 and a base 7.24 — 87 of 90 tasks take exactly the oracle number of
|
| 60 |
+
calls, 3 take one fewer, none take more.
|
| 61 |
- Tool error rate **0.00%** on both splits, loop collapse **0.00%**, redundant calls **0.00%**,
|
| 62 |
parse failures **0.00%** (base: 22.3% tool errors, 16 of 30 episodes never finishing,
|
| 63 |
`overcall` 1.08).
|