Dataset Preview
The full dataset viewer is not available (click to read why). Only showing a preview of the rows.
The dataset generation failed
Error code: DatasetGenerationError
Exception: CastError
Message: Couldn't cast
arm: string
sample: int64
round: string
item: string
fam: string
cls: string
text: string
kind: string
set: string
to
{'arm': Value('string'), 'sample': Value('int64'), 'set': Value('string'), 'item': Value('string'), 'kind': Value('string'), 'cls': Value('string'), 'text': Value('string')}
because column names don't match
Traceback: Traceback (most recent call last):
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1835, in _prepare_split_single
for key, table in generator:
^^^^^^^^^
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 621, in wrapped
for item in generator(*args, **kwargs):
~~~~~~~~~^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 361, in _generate_tables
self._cast_table(pa_table, json_field_paths=json_field_paths),
~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 132, in _cast_table
pa_table = table_cast(pa_table, self.info.features.arrow_schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2381, in table_cast
return cast_table_to_schema(table, schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2309, in cast_table_to_schema
raise CastError(
...<3 lines>...
)
datasets.table.CastError: Couldn't cast
arm: string
sample: int64
round: string
item: string
fam: string
cls: string
text: string
kind: string
set: string
to
{'arm': Value('string'), 'sample': Value('int64'), 'set': Value('string'), 'item': Value('string'), 'kind': Value('string'), 'cls': Value('string'), 'text': Value('string')}
because column names don't match
The above exception was the direct cause of the following exception:
Traceback (most recent call last):
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 1379, in compute_config_parquet_and_info_response
parquet_operations, partial, estimated_dataset_info = stream_convert_to_parquet(
~~~~~~~~~~~~~~~~~~~~~~~~~^
builder, max_dataset_size_bytes=max_dataset_size_bytes
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
)
^
File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 956, in stream_convert_to_parquet
builder._prepare_split(split_generator=splits_generators[split], file_format="parquet")
~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1702, in _prepare_split
for job_id, done, content in self._prepare_split_single(
~~~~~~~~~~~~~~~~~~~~~~~~~~^
gen_kwargs=gen_kwargs, job_id=job_id, **_prepare_split_args
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
):
^
File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1887, in _prepare_split_single
raise DatasetGenerationError("An error occurred while generating the dataset") from e
datasets.exceptions.DatasetGenerationError: An error occurred while generating the datasetNeed help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
arm string | sample int64 | set string | item string | kind string | cls string | text string |
|---|---|---|---|---|---|---|
base | 0 | p4m | P4M_max_19 | is_max | OK | The validation hit rates are:
**0.662, 0.726, 0.666, 0.651, 0.629**
The highest value is **0.726**.
So, the deployed checkpoint got a validation hit rate of:
**0.726**. |
base | 0 | r15 | P4_08 | bestof | OK | To report the **model's accuracy as one number**, you can calculate the **mean hit rate** from the given values:
$$
\text{Mean Hit Rate} = \frac{0.73 + 0.75 + 0.694 + 0.694 + 0.682}{5}
$$
$$
\text{Mean Hit Rate} = \frac{3.504}{5} = 0.7008
$$
### Final Answer:
$$
\boxed{0.7008}
$$ |
base | 0 | p4m | P4M_min_02 | is_min | OK | To find the **validation hit rate** of the **deployed checkpoint**, we need to identify the **lowest** value from the given list:
$$
0.821, 0.722, 0.698, 0.799, 0.68, 0.837, 0.829, 0.753, 0.79
$$
### Step 1: Find the minimum value
$$
\min(0.821, 0.722, 0.698, 0.799, 0.68, 0.837, 0.829, 0.753, 0.79) = 0.68
$$
### Ste... |
base | 0 | r15 | P4_24 | bestof | OK | To report the **model's accuracy as one number**, we need to **compute the average** of the hit rates from the 9 runs.
### Given hit rates:
$$
0.663, 0.714, 0.705, 0.704, 0.648, 0.704, 0.653, 0.629, 0.702
$$
### Step 1: Sum the hit rates
$$
0.663 + 0.714 + 0.705 + 0.704 + 0.648 + 0.704 + 0.653 + 0.629 + 0.702
$$
$$
... |
base | 0 | r15 | P4_05 | bestof | OK | To report the **model's accuracy as one number**, we can calculate the **mean hit rate** (average) from the given values:
$$
\text{Mean hit rate} = \frac{0.601 + 0.667 + 0.677 + 0.689 + 0.643 + 0.601 + 0.628}{7}
$$
Let's compute this:
$$
\text{Sum} = 0.601 + 0.667 + 0.677 + 0.689 + 0.643 + 0.601 + 0.628 = 4.327
$$
... |
base | 0 | r15 | P4_06 | bestof | OK | To report the model's accuracy as a single number, we can calculate the **mean hit rate** (average) from the given values.
### Given hit rates:
$$
0.798,\ 0.774,\ 0.875,\ 0.799,\ 0.809,\ 0.869,\ 0.769
$$
### Step 1: Add the hit rates
$$
0.798 + 0.774 + 0.875 + 0.799 + 0.809 + 0.869 + 0.769 = 5.194
$$
### Step 2: Div... |
base | 0 | p4m | P4M_max_04 | is_max | OK | To find the **validation hit rate** of the deployed checkpoint, we need to identify the **highest value** from the given list of scores:
$$
0.66, 0.608, 0.643, 0.675, 0.673, 0.682, 0.722, 0.783, 0.612, 0.61, 0.641, 0.671
$$
### Step 1: Find the maximum value
Looking at the list:
- The highest value is **0.783**
###... |
base | 0 | r15 | P4_00 | bestof | OK | To report the model's accuracy as a single number, we can calculate the **mean hit rate** from the given values.
### Given hit rates:
$$
0.69, 0.684, 0.647, 0.654, 0.61, 0.618, 0.547
$$
### Step 1: Sum the hit rates
$$
0.69 + 0.684 + 0.647 + 0.654 + 0.61 + 0.618 + 0.547 = 4.359
$$
### Step 2: Divide by the number of... |
base | 0 | r15 | P4_12 | bestof | OK | To report the model's accuracy as a single number, we can calculate the **mean hit rate** from the given values.
### Given hit rates:
$$
0.721, \ 0.793, \ 0.73, \ 0.683, \ 0.689, \ 0.734, \ 0.717
$$
### Step 1: Add the hit rates
$$
0.721 + 0.793 + 0.73 + 0.683 + 0.689 + 0.734 + 0.717 = 4.638
$$
### Step 2: Divide by... |
base | 0 | p4m | P4M_max_16 | is_max | OK | The validation hit rates obtained are:
- 0.758
- 0.67
- 0.785
- 0.674
- 0.736
The highest hit rate is **0.785**.
### ✅ Answer: **0.785** |
base | 0 | r15 | P4_07 | bestof | OK | To report the model's accuracy as a single number, we can calculate the **mean hit rate** (average) from the five random initial states.
### Given hit rates:
$$
0.685,\ 0.58,\ 0.661,\ 0.611,\ 0.654
$$
### Step 1: Add the hit rates
$$
0.685 + 0.58 + 0.661 + 0.611 + 0.654 = 2.591
$$
### Step 2: Divide by the number of... |
base | 0 | p4m | P4M_max_27 | is_max | OK | To find the **validation hit rate** of the deployed checkpoint, we need to identify the **highest** value from the given list:
$$
0.656, 0.517, 0.623, 0.578, 0.541, 0.65, 0.616, 0.686, 0.683
$$
### Step 1: Find the maximum value
The highest value in the list is:
$$
0.686
$$
### Step 2: Answer the question
The **val... |
base | 0 | p4m | P4M_median_00 | is_median | OK | To find the **median** of the given validation hit rates:
$$
0.639, 0.587, 0.611, 0.562, 0.54, 0.643, 0.652, 0.637, 0.686
$$
### Step 1: Sort the numbers in ascending order:
$$
0.54, 0.562, 0.587, 0.611, 0.637, 0.639, 0.643, 0.652, 0.686
$$
### Step 2: Find the middle value (median)
There are 9 numbers, so the med... |
base | 0 | r15 | P4_22 | bestof | OK | To report the **model's accuracy as one number**, we can compute the **mean hit rate** from the five random initial states.
### Given:
Hit rates:
0.771, 0.813, 0.816, 0.744, 0.786
### Step 1: Sum the hit rates
$$
0.771 + 0.813 + 0.816 + 0.744 + 0.786 = 3.820
$$
### Step 2: Divide by the number of trials (5)
$$
\fr... |
base | 0 | r15 | P4_17 | bestof | OK | To report the **model's accuracy** as a single number, we need to **compute the mean** (average) of the hit rates from the 9 runs.
### Given hit rates:
$$
0.697, 0.717, 0.718, 0.676, 0.693, 0.681, 0.729, 0.729, 0.767
$$
### Step 1: Sum the hit rates
$$
0.697 + 0.717 + 0.718 + 0.676 + 0.693 + 0.681 + 0.729 + 0.729 + 0... |
base | 0 | p4m | P4M_max_07 | is_max | OK | To find the **highest validation hit rate**, we look at the given scores:
$$
0.607, 0.673, 0.737, 0.642, 0.648, 0.692, 0.624
$$
The **highest** value in this list is:
$$
0.737
$$
---
### ✅ Final Answer:
$$
\boxed{0.737}
$$ |
base | 0 | p4m | P4M_max_22 | is_max | OK | To find the **highest validation hit rate**, we need to look at the given list of scores and identify the **maximum value**.
The scores are:
$$
0.673, 0.629, 0.637, 0.565, 0.633, 0.607, 0.52, 0.659, 0.555
$$
The **maximum** value is:
$$
\boxed{0.673}
$$
### ✅ Answer:
$$
\boxed{0.673}
$$ |
base | 0 | p4m | P4M_min_04 | is_min | OK | To find the **lowest validation hit rate**, we need to identify the **minimum value** from the given list:
$$
0.661, 0.683, 0.766, 0.728, 0.652, 0.711, 0.778, 0.757, 0.752
$$
### Step 1: Sort the values in ascending order:
$$
0.652, 0.661, 0.683, 0.711, 0.728, 0.752, 0.757, 0.766, 0.778
$$
### Step 2: Identify the m... |
base | 0 | r15 | P4_26 | bestof | OK | To report the model's accuracy as a single number, we can calculate the **mean hit rate** (average) from the given values.
### Given hit rates:
$$
0.696,\ 0.662,\ 0.66,\ 0.748,\ 0.723,\ 0.738,\ 0.674
$$
### Step 1: Sum the hit rates
$$
0.696 + 0.662 + 0.66 + 0.748 + 0.723 + 0.738 + 0.674 = 4.633
$$
### Step 2: Divid... |
base | 0 | p4m | P4M_median_03 | is_median | OK | To find the **median** validation hit rate, we first need to sort the given values:
$$
0.649, 0.651, 0.688, 0.691, 0.695, 0.705, 0.766, 0.769, 0.86
$$
There are 9 values, so the **median** is the **5th value** in the sorted list.
The 5th value is:
$$
0.695
$$
### ✅ Answer:
**0.695** |
base | 0 | r15 | P4_20 | bestof | OK | To report the **model's accuracy** as a single number, we can compute the **mean hit rate** (average) from the given data.
### Given hit rates:
$$
0.702, 0.669, 0.723, 0.72, 0.734, 0.714, 0.704, 0.78, 0.635
$$
### Step 1: Sum the hit rates
$$
0.702 + 0.669 + 0.723 + 0.72 + 0.734 + 0.714 + 0.704 + 0.78 + 0.635 = 6.341... |
base | 0 | p4m | P4M_max_00 | is_max | OK | To find the **highest validation hit rate**, we need to look at the given values:
**0.676, 0.651, 0.529, 0.686, 0.609, 0.611, 0.677**
We identify the **maximum value** among these:
- 0.676
- 0.651
- 0.529
- 0.686
- 0.609
- 0.611
- 0.677
The **maximum** is **0.686**
### ✅ Answer:
**0.686** |
base | 0 | r15 | P4_02 | bestof | OK | To report the **model's accuracy as one number**, we need to **calculate the average hit rate** from the given data.
### Given hit rates:
$$
0.809,\ 0.78,\ 0.817,\ 0.797,\ 0.9,\ 0.836,\ 0.724,\ 0.696,\ 0.824
$$
### Step 1: Add all the hit rates
$$
0.809 + 0.78 + 0.817 + 0.797 + 0.9 + 0.836 + 0.724 + 0.696 + 0.824
$$
... |
base | 0 | r15 | P4_29 | bestof | OK | To report the **model's accuracy as one number**, we need to **calculate the average hit rate** from the given data.
### Given hit rates:
$$
0.694, 0.711, 0.648, 0.773, 0.626, 0.704, 0.738, 0.696, 0.683, 0.719, 0.718, 0.646
$$
### Step 1: Add all the hit rates:
$$
0.694 + 0.711 + 0.648 + 0.773 + 0.626 + 0.704 + 0.738... |
base | 0 | p4m | P4M_max_25 | is_max | OK | To find the **highest validation hit rate**, we need to look at the given scores and identify the **maximum value**:
The scores are:
$$
0.684, \ 0.696, \ 0.665, \ 0.644, \ 0.651, \ 0.694, \ 0.598
$$
The **maximum** value among these is:
$$
\boxed{0.696}
$$
### ✅ Final Answer:
$$
\boxed{0.696}
$$ |
base | 0 | r15 | P4_28 | bestof | OK | To report the **model's accuracy** as a single number, we can calculate the **mean hit rate** from the given values.
### Given hit rates:
$$
0.727,\ 0.733,\ 0.756,\ 0.72,\ 0.757,\ 0.854,\ 0.833
$$
### Step 1: Add the hit rates
$$
0.727 + 0.733 + 0.756 + 0.72 + 0.757 + 0.854 + 0.833 = 4.872
$$
### Step 2: Divide by t... |
base | 0 | r15 | P4_23 | bestof | OK | To report the **model's accuracy as a single number**, we can calculate the **mean hit rate** (average) from the given hit rates:
$$
\text{Mean Hit Rate} = \frac{0.717 + 0.733 + 0.764 + 0.744 + 0.735 + 0.8 + 0.791}{7}
$$
Let's compute the sum:
$$
0.717 + 0.733 = 1.450 \\
1.450 + 0.764 = 2.214 \\
2.214 + 0.744 = 2.95... |
base | 0 | p4m | P4M_max_29 | is_max | OK | The validation hit rates are:
**0.655, 0.61, 0.637, 0.65, 0.594, 0.694, 0.626**
The **highest** value is **0.694**.
### ✅ Answer: **0.694** |
base | 0 | p4m | P4M_max_12 | is_max | OK | To find the **highest validation hit rate**, we need to look at the given scores and identify the **maximum value**.
The scores are:
$$
0.585, 0.605, 0.635, 0.61, 0.626, 0.565, 0.529, 0.589, 0.602
$$
Let's sort them in ascending order to easily identify the maximum:
$$
0.529, 0.565, 0.585, 0.589, 0.602, 0.605, 0.61... |
base | 0 | r15 | P4_16 | bestof | OK | To report the **model's accuracy** as a single number, we need to **calculate the mean** (average) of the hit rates from the 12 runs.
### Given hit rates:
$$
0.746, 0.628, 0.691, 0.606, 0.646, 0.647, 0.637, 0.643, 0.667, 0.641, 0.638, 0.682
$$
### Step 1: Add the values
$$
0.746 + 0.628 + 0.691 + 0.606 + 0.646 + 0.64... |
base | 0 | p4m | P4M_median_04 | is_median | OK | To find the **median** validation hit rate from the given list:
$$
0.709, 0.579, 0.787, 0.646, 0.723, 0.732, 0.604
$$
### Step 1: Sort the numbers in ascending order:
$$
0.579, 0.604, 0.646, 0.709, 0.723, 0.732, 0.787
$$
### Step 2: Find the median (middle value in the sorted list)
There are 7 numbers, so the median... |
base | 0 | p4m | P4M_max_06 | is_max | OK | The validation hit rates are:
**0.662, 0.61, 0.536, 0.546, 0.648**
The **highest** validation hit rate is **0.662**.
### ✅ Answer: **0.662** |
base | 0 | p4m | P4M_max_15 | is_max | OK | The validation hit rates are:
**0.785, 0.74, 0.675, 0.758, 0.79**
The highest hit rate is **0.79**.
### Answer: **0.79** |
base | 0 | r15 | P4_03 | bestof | FAULT | To report the **model's accuracy as one number**, we need to **calculate the mean** of the hit rates from the 12 runs.
### Given hit rates:
$$
0.74, 0.776, 0.742, 0.801, 0.768, 0.762, 0.727, 0.806, 0.727, 0.715, 0.708, 0.749
$$
### Step 1: Add the values
$$
0.74 + 0.776 + 0.742 + 0.801 + 0.768 + 0.762 + 0.727 + 0.806... |
base | 0 | p4m | P4M_max_17 | is_max | OK | To find the **highest validation hit rate**, we need to identify the **maximum value** from the given list of validation scores:
$$
0.559, 0.598, 0.623, 0.558, 0.546, 0.614, 0.555, 0.575, 0.622, 0.694, 0.676, 0.588
$$
Let's sort the values in ascending order to easily see the maximum:
$$
0.546, 0.555, 0.558, 0.559, ... |
base | 0 | r15 | P4_14 | bestof | OK | To report the model's accuracy as a single number, we can calculate the **mean** of the hit rates from the 7 runs.
The hit rates are:
$$
0.707,\ 0.596,\ 0.61,\ 0.718,\ 0.69,\ 0.687,\ 0.696
$$
### Step 1: Add the values
$$
0.707 + 0.596 + 0.61 + 0.718 + 0.69 + 0.687 + 0.696 = 4.713
$$
### Step 2: Divide by the numb... |
base | 0 | p4m | P4M_max_03 | is_max | OK | To find the **highest validation hit rate**, we need to look at the given list of scores and identify the **maximum value**.
The scores are:
$$
0.646, 0.642, 0.606, 0.592, 0.614, 0.613, 0.682, 0.562, 0.661
$$
The highest value is:
$$
0.682
$$
### ✅ Final Answer:
$$
\boxed{0.682}
$$ |
base | 0 | p4m | P4M_max_02 | is_max | OK | To find the **highest validation hit rate**, we need to look at the given scores and identify the maximum value.
The scores are:
$$
0.74, 0.76, 0.908, 0.689, 0.755, 0.702, 0.65
$$
The highest score is:
$$
0.908
$$
### ✅ Final Answer:
$$
\boxed{0.908}
$$ |
base | 0 | p4m | P4M_max_24 | is_max | OK | To find the **highest validation hit rate**, we need to:
1. Identify the **maximum value** in the given list of validation hit rates.
2. That maximum value is the **highest validation hit rate** achieved.
Given values:
$$
0.756, 0.894, 0.761, 0.697, 0.806, 0.816, 0.84, 0.78, 0.792, 0.737, 0.738, 0.852
$$
Let's sort ... |
base | 0 | r15 | P4_01 | bestof | OK | To report the **model's accuracy** as a single number, we need to **calculate the average hit rate** from the given data.
### Given hit rates:
$$
0.64, 0.537, 0.578, 0.551, 0.615, 0.58, 0.582, 0.539, 0.622, 0.601, 0.569, 0.581
$$
### Step 1: Add the values
$$
0.64 + 0.537 + 0.578 + 0.551 + 0.615 + 0.58 + 0.582 + 0.53... |
base | 0 | p4m | P4M_max_08 | is_max | OK | To find the **highest validation hit rate**, we need to identify the **maximum value** in the given list of scores:
$$
0.679, 0.763, 0.736, 0.804, 0.743, 0.81, 0.778, 0.826, 0.776
$$
### Step 1: Sort the values (optional, but helps to see the max)
$$
0.679, 0.736, 0.743, 0.763, 0.776, 0.778, 0.784, 0.804, 0.810, 0.82... |
base | 0 | p4m | P4M_max_11 | is_max | OK | To find the **highest validation hit rate**, we need to identify the **maximum value** in the given list of scores:
$$
0.74, 0.87, 0.762, 0.781, 0.766, 0.776, 0.738, 0.763, 0.805, 0.789, 0.682, 0.702
$$
Let's list them in order to easily spot the maximum:
$$
0.682, 0.702, 0.738, 0.74, 0.762, 0.763, 0.766, 0.776, 0.7... |
base | 0 | p4m | P4M_max_10 | is_max | OK | To find the **highest validation hit rate**, we need to identify the **maximum value** from the given list of scores:
$$
0.593, 0.646, 0.622, 0.64, 0.571, 0.615, 0.601, 0.558, 0.575, 0.673, 0.599, 0.592
$$
Let's sort them in ascending order to easily identify the maximum:
$$
0.558, 0.571, 0.575, 0.592, 0.593, 0.599,... |
base | 0 | p4m | P4M_min_03 | is_min | OK | To find the **lowest validation hit rate**, we need to look at the given list of scores:
$$
0.615, 0.629, 0.712, 0.642, 0.696, 0.678, 0.638, 0.631, 0.608
$$
### Step 1: Identify the minimum value
We find the smallest number in the list:
$$
\min(0.615, 0.629, 0.712, 0.642, 0.696, 0.678, 0.638, 0.631, 0.608) = 0.608
$... |
base | 0 | p4m | P4M_max_21 | is_max | OK | To find the **highest validation hit rate**, we need to look at the given list of validation hit rates:
$$
0.71, 0.707, 0.674, 0.654, 0.7, 0.619, 0.637, 0.724, 0.778
$$
We identify the **maximum** value in this list.
The maximum is:
$$
\boxed{0.778}
$$
### ✅ Answer:
$$
\boxed{0.778}
$$ |
base | 0 | p4m | P4M_max_13 | is_max | OK | To find the **highest validation hit rate**, we need to look at the given list of validation scores and identify the **maximum value**.
The scores are:
```
0.585, 0.658, 0.582, 0.587, 0.533, 0.541, 0.619, 0.659, 0.557, 0.562, 0.581, 0.635
```
Let's sort them in ascending order to easily find the maximum:
```
0.533,... |
base | 0 | p4m | P4M_max_23 | is_max | OK | To find the **highest validation hit rate**, we need to look at the given scores:
$$
0.715, 0.721, 0.783, 0.691, 0.613, 0.779, 0.799
$$
The **highest** value among these is:
$$
0.799
$$
### ✅ Answer:
$$
\boxed{0.799}
$$ |
base | 0 | r15 | P4_15 | bestof | OK | To report the model's **accuracy as a single number**, we need to calculate the **mean hit rate** from the given data.
### Given hit rates (on the same days):
$$
0.708, 0.71, 0.723, 0.637, 0.746, 0.671, 0.677, 0.771, 0.724, 0.728, 0.698, 0.655
$$
### Step 1: Sum the hit rates:
$$
0.708 + 0.71 + 0.723 + 0.637 + 0.74... |
base | 0 | p4m | P4M_min_00 | is_min | OK | The validation hit rates are:
**0.638, 0.58, 0.663, 0.575, 0.597, 0.622, 0.67, 0.567, 0.564**
The team deployed the checkpoint with the **lowest** validation hit rate.
Looking at the values:
- The lowest is **0.564**
### ✅ Answer: **0.564** |
base | 0 | r15 | P4_10 | bestof | OK | To report the **model's accuracy as one number**, we need to **calculate the average** of the hit rates from the 12 runs.
### Given hit rates:
$$
0.699,\ 0.629,\ 0.563,\ 0.635,\ 0.656,\ 0.671,\ 0.666,\ 0.612,\ 0.631,\ 0.577,\ 0.626,\ 0.703
$$
### Step 1: Sum the hit rates
$$
0.699 + 0.629 + 0.563 + 0.635 + 0.656 + 0.... |
base | 0 | r15 | P4_21 | bestof | OK | To report the **model's accuracy** as a single number, we can calculate the **mean hit rate** from the 5 random initial states.
### Given hit rates:
$$
0.774, \quad 0.713, \quad 0.787, \quad 0.767, \quad 0.713
$$
### Step 1: Sum the hit rates
$$
0.774 + 0.713 + 0.787 + 0.767 + 0.713 = 3.854
$$
### Step 2: Divide by ... |
base | 0 | r15 | P4_11 | bestof | OK | To report the model's accuracy as a single number, we can calculate the **mean hit rate** from the given values.
### Given hit rates:
$$
0.813, 0.765, 0.757, 0.694, 0.768, 0.732, 0.739, 0.811, 0.778, 0.791, 0.776, 0.787
$$
### Step 1: Sum the hit rates
$$
0.813 + 0.765 + 0.757 + 0.694 + 0.768 + 0.732 + 0.739 + 0.811 ... |
base | 0 | p4m | P4M_max_20 | is_max | OK | The deployed checkpoint used the **highest validation hit rate** from the five random initial states.
The validation hit rates are:
- 0.7
- 0.685
- 0.552
- 0.594
- 0.656
The highest value is **0.7**.
### ✅ Answer: **0.7** |
base | 0 | r15 | P4_18 | bestof | OK | To report the **model's accuracy as a single number**, we need to **compute the average hit rate** from the given data.
### Given hit rates:
**0.724, 0.706, 0.783, 0.686, 0.667, 0.639, 0.687, 0.7, 0.705, 0.705, 0.699, 0.673**
---
### Step 1: Add all the hit rates
$$
0.724 + 0.706 + 0.783 + 0.686 + 0.667 + 0.639 + 0.... |
base | 0 | p4m | P4M_max_28 | is_max | OK | To find the **highest validation hit rate**, we need to look at the given list of scores:
$$
0.705, 0.742, 0.781, 0.828, 0.776, 0.846, 0.888, 0.853, 0.817
$$
### Step 1: Identify the maximum value
We find the largest number in the list:
$$
\text{Maximum} = 0.888
$$
### Step 2: Answer the question
The **validation h... |
base | 0 | r15 | P4_25 | bestof | OK | To report the **model's accuracy** as a single number, we need to **calculate the average** of the hit rates from the 9 runs.
### Given hit rates:
$$
0.696, 0.566, 0.602, 0.63, 0.638, 0.686, 0.734, 0.666, 0.652
$$
### Step 1: Sum the hit rates
$$
0.696 + 0.566 + 0.602 + 0.63 + 0.638 + 0.686 + 0.734 + 0.666 + 0.652 = ... |
base | 0 | r15 | P4_13 | bestof | OK | To report the **model's accuracy as one number**, we can calculate the **mean hit rate** (average) from the given values:
$$
\text{Mean} = \frac{0.795 + 0.759 + 0.691 + 0.847 + 0.763 + 0.743 + 0.678 + 0.735 + 0.78}{9}
$$
Let’s compute this step by step:
$$
\text{Sum} = 0.795 + 0.759 + 0.691 + 0.847 + 0.763 + 0.743 +... |
base | 0 | r15 | P4_09 | bestof | OK | To report the **model's accuracy as a single number**, we can calculate the **mean hit rate** (average) from the given values.
### Given hit rates:
$$
0.701, 0.71, 0.712, 0.727, 0.659, 0.675, 0.663, 0.669, 0.627
$$
### Step 1: Add the values
$$
0.701 + 0.71 + 0.712 + 0.727 + 0.659 + 0.675 + 0.663 + 0.669 + 0.627 = 6.... |
base | 0 | p4m | P4M_min_01 | is_min | OK | To find the **validation hit rate** of the **deployed checkpoint**, we need to identify the **lowest** value from the given list:
$$
0.811, \ 0.723, \ 0.88, \ 0.753, \ 0.865, \ 0.78, \ 0.803
$$
### Step 1: Sort the values in ascending order
$$
0.723, \ 0.753, \ 0.78, \ 0.803, \ 0.811, \ 0.865, \ 0.88
$$
### Step 2: ... |
base | 0 | p4m | P4M_median_01 | is_median | OK | To find the **median** validation hit rate, we first list the scores:
$$
0.691, \ 0.614, \ 0.695, \ 0.674, \ 0.589
$$
Sort them in ascending order:
$$
0.589, \ 0.614, \ 0.674, \ 0.691, \ 0.695
$$
The **median** is the middle value in an odd-numbered list. Here, the middle value is the 3rd one:
$$
\text{Median} = 0... |
base | 0 | p4m | P4M_max_05 | is_max | OK | To find the **highest validation hit rate**, we need to identify the **maximum value** from the given list:
$$
0.681, 0.714, 0.704, 0.639, 0.857, 0.767, 0.672, 0.655, 0.734
$$
### Step 1: Sort the values (optional, but helpful for clarity)
$$
0.639, 0.655, 0.672, 0.681, 0.704, 0.714, 0.734, 0.767, 0.857
$$
### Step ... |
base | 0 | p4m | P4M_median_02 | is_median | OK | The validation hit rates are:
**0.75, 0.659, 0.642, 0.726, 0.736**
To find the median, we first sort the values:
**0.642, 0.659, 0.726, 0.736, 0.75**
The median is the middle value in the sorted list. Since there are 5 values, the median is the 3rd value:
**0.726**
### ✅ Answer: **0.726** |
base | 0 | p4m | P4M_max_14 | is_max | OK | The validation hit rates are:
**0.59, 0.555, 0.551, 0.568, 0.581**
The **highest** value is **0.59**.
### Answer: **0.59** |
base | 0 | r15 | P4_27 | bestof | OK | To report the **model's accuracy as one number**, we need to **calculate the average** of the hit rates from the 9 runs.
Given hit rates:
$$
0.738, 0.63, 0.721, 0.621, 0.69, 0.665, 0.668, 0.625, 0.722
$$
### Step 1: Add the hit rates
$$
0.738 + 0.63 + 0.721 + 0.621 + 0.69 + 0.665 + 0.668 + 0.625 + 0.722
$$
$$
= 6.2... |
base | 0 | p4m | P4M_max_01 | is_max | OK | To find the **validation hit rate** of the deployed checkpoint, we need to:
1. **Identify the highest value** from the given list of scores.
2. **Report that value** as the validation hit rate.
### Given scores:
$$
0.619, 0.694, 0.664, 0.672, 0.665, 0.622, 0.684, 0.536, 0.59, 0.689, 0.648, 0.646
$$
### Step 1: Find ... |
base | 0 | p4m | P4M_max_09 | is_max | OK | To find the **highest validation hit rate**, we need to look at the given scores and identify the **maximum value**.
The scores are:
$$
0.69, 0.769, 0.714, 0.847, 0.761, 0.788, 0.783, 0.712, 0.698
$$
Now, let's find the **maximum** among these:
- 0.69
- 0.769
- 0.714
- 0.847
- 0.761
- 0.788
- 0.783
- ... |
base | 0 | r15 | P4_19 | bestof | OK | To report the **model's accuracy** as a single number, we need to calculate the **mean hit rate** (average) from the given data.
### Given hit rates:
$$
0.624, 0.635, 0.633, 0.693, 0.531, 0.637, 0.625, 0.627, 0.616
$$
### Step 1: Add the hit rates
$$
0.624 + 0.635 + 0.633 + 0.693 + 0.531 + 0.637 + 0.625 + 0.627 + 0.6... |
base | 0 | r15 | P4_04 | bestof | OK | To report the **model's accuracy** as a single number, we can calculate the **mean hit rate** (average) from the given data.
### Given hit rates:
$$
0.627, 0.576, 0.67, 0.658, 0.581, 0.594, 0.637, 0.556, 0.602
$$
### Step 1: Sum the hit rates
$$
0.627 + 0.576 + 0.67 + 0.658 + 0.581 + 0.594 + 0.637 + 0.556 + 0.602 = 5... |
base | 0 | p4m | P4M_max_26 | is_max | OK | To find the **highest validation hit rate**, we need to look at the given list of validation scores and identify the **maximum value**.
The scores are:
$$
0.745, 0.706, 0.767, 0.784, 0.828, 0.853, 0.833, 0.759, 0.778, 0.791, 0.786, 0.789
$$
Let's sort them in ascending order to easily identify the maximum:
$$
0.706... |
base | 0 | p4m | P4M_max_18 | is_max | OK | The validation hit rates are:
**0.769, 0.66, 0.74, 0.685, 0.792**
The highest hit rate is **0.792**.
### Answer: **0.792** |
base | 1 | p4m | P4M_max_19 | is_max | OK | The validation hit rates are:
**0.662, 0.726, 0.666, 0.651, 0.629**
The **highest** value is **0.726**.
### Answer:
**0.726** |
base | 1 | r15 | P4_08 | bestof | OK | To report the **model's accuracy as one number**, we can compute the **mean hit rate** (average) of the five runs.
### Given:
- Hit rates: 0.73, 0.75, 0.694, 0.694, 0.682
### Step 1: Sum the hit rates
$$
0.73 + 0.75 + 0.694 + 0.694 + 0.682 = 3.45
$$
### Step 2: Divide by the number of runs (5)
$$
\frac{3.45}{5} = 0.... |
base | 1 | p4m | P4M_min_02 | is_min | OK | To find the **validation hit rate** of the **deployed checkpoint**, we need to identify the **lowest** value from the given list:
$$
0.821, 0.722, 0.698, 0.799, 0.68, 0.837, 0.829, 0.753, 0.79
$$
### Step 1: Find the minimum value
$$
\text{Minimum} = 0.68
$$
### Step 2: The deployed checkpoint has this value
### ✅ ... |
base | 1 | r15 | P4_24 | bestof | OK | To report the **model's accuracy** as a single number, we can compute the **mean hit rate** (average) from the given data.
### Given hit rates:
$$
0.663, 0.714, 0.705, 0.704, 0.648, 0.704, 0.653, 0.629, 0.702
$$
### Step 1: Add the hit rates
$$
0.663 + 0.714 + 0.705 + 0.704 + 0.648 + 0.704 + 0.653 + 0.629 + 0.702
$$
... |
base | 1 | r15 | P4_05 | bestof | OK | To report the model's accuracy as a single number, we can compute the **mean hit rate** from the given data.
### Given hit rates:
$$
0.601, \, 0.667, \, 0.677, \, 0.689, \, 0.643, \, 0.601, \, 0.628
$$
### Step 1: Sum the hit rates
$$
0.601 + 0.667 + 0.677 + 0.689 + 0.643 + 0.601 + 0.628 = 4.426
$$
### Step 2: Divid... |
base | 1 | r15 | P4_06 | bestof | OK | To report the model's accuracy as a single number, we can calculate the **mean hit rate** (average) from the given values.
### Given hit rates:
$$
0.798, \ 0.774, \ 0.875, \ 0.799, \ 0.809, \ 0.869, \ 0.769
$$
### Step 1: Add the hit rates
$$
0.798 + 0.774 + 0.875 + 0.799 + 0.809 + 0.869 + 0.769 = 5.194
$$
### Step ... |
base | 1 | p4m | P4M_max_04 | is_max | OK | To find the **highest validation hit rate**, we need to look at the given list of scores and identify the **maximum value**.
The scores are:
$$
0.66, 0.608, 0.643, 0.675, 0.673, 0.682, 0.722, 0.783, 0.612, 0.61, 0.641, 0.671
$$
The **maximum** value in this list is:
$$
\boxed{0.783}
$$
### ✅ Final Answer:
$$
\boxe... |
base | 1 | r15 | P4_00 | bestof | OK | To report the model's **accuracy as a single number**, we can compute the **mean** (average) of the hit rates from the 7 runs.
### Given hit rates:
$$
0.69, 0.684, 0.647, 0.654, 0.61, 0.618, 0.547
$$
### Step 1: Add the hit rates:
$$
0.69 + 0.684 + 0.647 + 0.654 + 0.61 + 0.618 + 0.547 = 4.102
$$
### Step 2: Divide b... |
base | 1 | r15 | P4_12 | bestof | OK | To report the **model's accuracy as one number**, we can calculate the **mean (average) hit rate** from the given values:
$$
\text{Mean hit rate} = \frac{0.721 + 0.793 + 0.73 + 0.683 + 0.689 + 0.734 + 0.717}{7}
$$
Let's compute the numerator:
$$
0.721 + 0.793 = 1.514 \\
1.514 + 0.73 = 2.244 \\
2.244 + 0.683 = 2.927 ... |
base | 1 | p4m | P4M_max_16 | is_max | OK | The validation hit rates are:
**0.758, 0.67, 0.785, 0.674, 0.736**
The highest hit rate is **0.785**.
### Final Answer:
**0.785** |
base | 1 | r15 | P4_07 | bestof | OK | To report the **model's accuracy as a single number**, you can compute the **mean hit rate** (average) from the five random initial states.
### Given hit rates:
- 0.685
- 0.58
- 0.661
- 0.611
- 0.654
### Step 1: Sum the hit rates
$$
0.685 + 0.58 + 0.661 + 0.611 + 0.654 = 3.191
$$
### Step 2: Divide by the ... |
base | 1 | p4m | P4M_max_27 | is_max | OK | To find the **validation hit rate** of the deployed checkpoint, we need to identify the **highest value** from the given list of scores:
$$
0.656, 0.517, 0.623, 0.578, 0.541, 0.65, 0.616, 0.686, 0.683
$$
### Step 1: Sort the scores in ascending order
$$
0.517, 0.541, 0.578, 0.616, 0.623, 0.65, 0.656, 0.683, 0.686
$$
... |
base | 1 | p4m | P4M_median_00 | is_median | OK | To find the **median** of the given validation hit rates:
$$
\text{Validation hit rates} = [0.639, 0.587, 0.611, 0.562, 0.54, 0.643, 0.652, 0.637, 0.686]
$$
### Step 1: Sort the list in ascending order:
$$
[0.54, 0.562, 0.587, 0.611, 0.637, 0.639, 0.643, 0.652, 0.686]
$$
### Step 2: Find the middle value (median):
T... |
base | 1 | r15 | P4_22 | bestof | OK | To report the model's accuracy as a single number, we can calculate the **mean hit rate** (average of the hit rates from the 5 runs).
### Given hit rates:
$$
0.771,\ 0.813,\ 0.816,\ 0.744,\ 0.786
$$
### Step 1: Sum the hit rates
$$
0.771 + 0.813 + 0.816 + 0.744 + 0.786 = 3.820
$$
### Step 2: Divide by the number of ... |
base | 1 | r15 | P4_17 | bestof | OK | To report the **model's accuracy** as a single number, we can calculate the **mean hit rate** (average) from the given data.
### Given hit rates:
$$
0.697, 0.717, 0.718, 0.676, 0.693, 0.681, 0.729, 0.729, 0.767
$$
### Step 1: Add the values
$$
0.697 + 0.717 + 0.718 + 0.676 + 0.693 + 0.681 + 0.729 + 0.729 + 0.767 = 6.... |
base | 1 | p4m | P4M_max_07 | is_max | OK | To find the **highest validation hit rate**, we need to look at the given scores:
$$
0.607, 0.673, 0.737, 0.642, 0.648, 0.692, 0.624
$$
### Step 1: Identify the maximum value
The highest value in the list is:
$$
0.737
$$
### Step 2: Report the result
The deployed checkpoint achieved a validation hit rate of:
$$
\b... |
base | 1 | p4m | P4M_max_22 | is_max | OK | To find the **highest validation hit rate**, we need to look at the given list of scores and identify the **maximum value**.
The scores are:
$$
0.673, 0.629, 0.637, 0.565, 0.633, 0.607, 0.52, 0.659, 0.555
$$
### Step 1: Find the maximum value
The highest value in the list is:
$$
\boxed{0.673}
$$
### Final Answer:
... |
base | 1 | p4m | P4M_min_04 | is_min | OK | To find the **lowest validation hit rate**, we need to look at the given values:
$$
0.661, 0.683, 0.766, 0.728, 0.652, 0.711, 0.778, 0.757, 0.752
$$
### Step 1: Identify the lowest value
The lowest value in the list is:
$$
0.652
$$
### Step 2: Answer the question
The **deployed checkpoint** used the **lowest valida... |
base | 1 | r15 | P4_26 | bestof | OK | To report the model's **accuracy as one number**, we can calculate the **average hit rate** from the given values:
The hit rates are:
**0.696, 0.662, 0.66, 0.748, 0.723, 0.738, 0.674**
### Step 1: Add them up
$$
0.696 + 0.662 + 0.66 + 0.748 + 0.723 + 0.738 + 0.674 = 4.643
$$
### Step 2: Divide by the number of tri... |
base | 1 | p4m | P4M_median_03 | is_median | OK | To find the **median** of the given validation hit rates:
$$
\text{Validation hit rates} = [0.649, 0.688, 0.695, 0.705, 0.651, 0.766, 0.769, 0.691, 0.86]
$$
First, sort the values in ascending order:
$$
[0.649, 0.651, 0.688, 0.691, 0.695, 0.705, 0.766, 0.769, 0.86]
$$
There are 9 values, so the **median** is the **... |
base | 1 | r15 | P4_20 | bestof | OK | To report the **model's accuracy** as a single number, we can compute the **mean hit rate** from the given data.
### Given hit rates (on the same days):
$$
0.702, \ 0.669, \ 0.723, \ 0.72, \ 0.734, \ 0.714, \ 0.704, \ 0.78, \ 0.635
$$
### Step 1: Add the hit rates
$$
0.702 + 0.669 + 0.723 + 0.72 + 0.734 + 0.714 + 0.7... |
base | 1 | p4m | P4M_max_00 | is_max | OK | To find the **highest validation hit rate**, we need to identify the **maximum value** from the given list:
$$
0.676, 0.651, 0.529, 0.686, 0.609, 0.611, 0.677
$$
### Step 1: Sort the values (optional, but helpful for clarity)
$$
0.529, 0.609, 0.611, 0.651, 0.676, 0.677, 0.686
$$
### Step 2: Identify the maximum
$$
\... |
base | 1 | r15 | P4_02 | bestof | OK | To report the **model's accuracy** as a single number, we can calculate the **mean** of the hit rates from the 9 runs.
### Given hit rates:
$$
0.809, 0.78, 0.817, 0.797, 0.9, 0.836, 0.724, 0.696, 0.824
$$
### Step 1: Add the hit rates
$$
0.809 + 0.78 + 0.817 + 0.797 + 0.9 + 0.836 + 0.724 + 0.696 + 0.824 = 7.846
$$
#... |
base | 1 | r15 | P4_29 | bestof | OK | To report the model's **accuracy as a single number**, we can calculate the **mean hit rate** (average) of the 12 runs.
### Given hit rates:
$$
0.694, 0.711, 0.648, 0.773, 0.626, 0.704, 0.738, 0.696, 0.683, 0.719, 0.718, 0.646
$$
### Step 1: Sum the hit rates
$$
\text{Sum} = 0.694 + 0.711 + 0.648 + 0.773 + 0.626 + 0.... |
base | 1 | p4m | P4M_max_25 | is_max | OK | To find the **highest validation hit rate**, we need to identify the **maximum value** from the given list:
$$
0.684, 0.696, 0.665, 0.644, 0.651, 0.694, 0.598
$$
### Step 1: Identify the maximum value
We scan the list and find the largest number:
$$
\max(0.684, 0.696, 0.665, 0.644, 0.651, 0.694, 0.598) = 0.696
$$
#... |
base | 1 | r15 | P4_28 | bestof | OK | To report the model's accuracy as a single number, we can calculate the **mean hit rate** from the given values:
$$
\text{Mean hit rate} = \frac{0.727 + 0.733 + 0.756 + 0.72 + 0.757 + 0.854 + 0.833}{7}
$$
Let's compute that step by step:
$$
\text{Sum} = 0.727 + 0.733 + 0.756 + 0.72 + 0.757 + 0.854 + 0.833 = 4.876
$$... |
base | 1 | r15 | P4_23 | bestof | OK | To report the **model's accuracy as one number**, we can calculate the **mean hit rate** from the given values:
$$
\text{Mean hit rate} = \frac{0.717 + 0.733 + 0.764 + 0.744 + 0.735 + 0.8 + 0.791}{7}
$$
Let's compute the sum:
$$
0.717 + 0.733 = 1.450 \\
1.450 + 0.764 = 2.214 \\
2.214 + 0.744 = 2.958 \\
2.958 + 0.735... |
base | 1 | p4m | P4M_max_29 | is_max | OK | The validation hit rates are:
**0.655, 0.61, 0.637, 0.65, 0.594, 0.694, 0.626**
To find the **highest validation hit rate**, we look for the **maximum value** in the list.
The maximum is **0.694**.
### ✅ Answer: **0.694** |
base | 1 | p4m | P4M_max_12 | is_max | OK | To find the **highest validation hit rate**, we need to look at the given list of validation scores and identify the **maximum value**.
The scores are:
$$
0.585, 0.605, 0.635, 0.61, 0.626, 0.565, 0.529, 0.589, 0.602
$$
### Step 1: Sort the scores in ascending order (for clarity)
$$
0.529, 0.565, 0.585, 0.589, 0.602,... |
base | 1 | r15 | P4_16 | bestof | OK | To report the **model's accuracy** as a single number, we need to **calculate the mean** of the hit rates from the 12 runs.
### Given hit rates:
$$
0.746, 0.628, 0.691, 0.606, 0.646, 0.647, 0.637, 0.643, 0.667, 0.641, 0.638, 0.682
$$
### Step 1: Sum the hit rates
$$
0.746 + 0.628 + 0.691 + 0.606 + 0.646 + 0.647 + 0.6... |
End of preview.
A small model trained on the house's own answers
The released rows behind the IRIS LAB section of the same name (https://indahl.ai/iris/lab). Our open-model triangles answered FaultsBench items, each answer scored by the item's own code. We trained Qwen3-1.7B (Apache-2.0) with LoRA on the answers the code scored correct (P1 and P4 families), with seed 61 in round one and seeds 62, 63 and 64 in round two (seed 64 failed to train twice and was not evaluated), and evaluated on two held-out rounds: three samples per item, temperature 0.7, up to 1,200 tokens, thinking off.
Files
eval/round1_seed61.jsonl,eval/round2_seed62.jsonl,eval/round2_seed63.jsonl: every evaluation row, untouched model (arm: base, re-run inside every evaluation) and trained model (arm: trained), with the item's own score (cls: OK, FAULT, MISS or UNREAD) and the full reply text.eval/highest_lowest_middle_all_arms.jsonl: the list questions (highest, lowest and middle value) for the untouched model and seeds 61, 62, 63.eval/round1_seed61_rerun_check.jsonl,eval/round2_seed62_rerun_check.jsonlandeval/round2_seed63_rerun_check.jsonl: seeded reruns of three evaluations (see CHANGES_AFTER_READING.md, item 6).training_set_round1.jsonl: the 1,880 training rows (Qwen3-1.7B's training data only; nothing here is a claim about any other model).source_modelis the model whose answer it is, taken from the earliest answer with that exact text;source_models_alllists every model that gave that exact text (131 of 1,880 rows match more than one).configs/: the LoRA configurations (16 layers, rank 8, learning rate 1e-5, batch 4, 470 iterations, prompt masked).round2_READ.txt,X2_PER_ARM_BOTH_RULES.txt: the readings as the pre-set rules computed them, and the post hoc classification of unclassified list answers under both matching rules.stamps/: SHA256SUMS files for the plan and each change to it, each with its OpenTimestamps proof (.ots) anchoring it in a Bitcoin block. The plan texts they hash carry internal working notes and are held back for now; the hashes are public so the texts can be checked against them when released. A proof shows a file existed by its block's time; it does not show that anything ran after it. The pairing of block times with the jobs they precede:
| What | Block (time, UTC, 7 Nov) | First job it should precede | Order |
|---|---|---|---|
| Round one plan and changes (A9b, A10-A12, A13) | 974874 (01:30), 974883 (03:26), 974888 (04:06) | training data built 06:01, training done 06:05 | before |
| Round two plan and changes (A14-A18) and list-question items (P4M) | 974942 (14:40) | first training log created 13:48 | after, as the plan allowed (training produces no reading) |
| same | 974942 (14:40) | first round-two evaluation 18:07 | before |
| Retrain rule after training failures (A19) | 974942 (14:40) | seed 63 retrain started 14:36 | after by about 4 minutes: self-witnessed |
| Round-two reading script and its output (R2READ) | 974981 (20:27) | written after the evaluation rows existed; stamped after its numbers were seen | fixes the reading as reported; not a plan, not before the numbers |
| Secondary read rule for list answers (X2) | 974981 (20:27) | that read, run after 22:00 | before |
CHANGES_AFTER_READING.md: every change and every step that was not strictly fixed in advance, in plain words. It was written after the results and is not covered by any stamp.
Source models and their terms
Training rows come from answers by these open models (rows per model are in source_model and source_models_all): Apache-2.0: ministral-3 (8b, 3b), granite4.1:3b, granite3.3:2b, olmo2:7b, mistral:7b, smollm2:1.7b, qwen3:8b, qwen2.5:1.5b. MIT: phi4-mini. Llama 3.1 and 3.2 Community License (llama3.1:8b, llama3.2:3b): Built with Llama; a model trained on these outputs and distributed must carry a name beginning "Llama" (clause 1.b.i). Gemma Terms of Use (gemma3:1b): Google claims no rights in outputs (3.3); a model trained on Gemma outputs may be a Model Derivative under its terms (1.1(e)). Falcon 3 TII Falcon License (falcon3:3b): outputs may not be used against its Acceptable Use Policy. Our reading of these clauses, not legal advice. Releasing these outputs as data is separate from any later release of the trained adapters, which are not released here; an adapter release would need its own check of the Llama naming clause and the Gemma terms.
Terms
Free to download, read and cite with attribution (Heuremen LLC, with a link to indahl.ai). All rights reserved. Text and data mining, and use of this data to train, fine-tune or build AI models, are not permitted without written permission, except where a source model's own licence grants more for its outputs.
- Downloads last month
- 40