Card: describe the benchmark through PyBot
Browse files- README.md +9 -18
- scripts/create_dataset.py +9 -17
README.md
CHANGED
|
@@ -61,14 +61,11 @@ instructions so that results compare across them.
|
|
| 61 |
|
| 62 |
## What it's for
|
| 63 |
|
| 64 |
-
SmolTalkBench
|
| 65 |
-
|
| 66 |
-
|
| 67 |
-
|
| 68 |
-
|
| 69 |
-
LMs, against the same fixed references. The best of these bots is
|
| 70 |
-
[PyBot](https://github.com/huggingface/pybot): its win rate climbed from 0% to 26% over more
|
| 71 |
-
than 6,000 commits, on par with SmolLM-360M-Instruct.
|
| 72 |
|
| 73 |

|
| 74 |
|
|
@@ -77,17 +74,14 @@ baselines' scores are inflated.
|
|
| 77 |
|
| 78 |
## How it's scored
|
| 79 |
|
| 80 |
-
The eval is the Inspect task `smoltalk_bench_gold` in flock's `experiments/run4/eval`, which
|
| 81 |
-
pins this dataset's `gold` config to a fixed revision.
|
| 82 |
-
|
| 83 |
- The model answers each `instruction` once, as a single user turn.
|
| 84 |
- An LLM judge (`deepseek-ai/DeepSeek-V4.1-Flash`, with AlpacaEval 2.0's weighted annotator
|
| 85 |
prompt) compares the reply with the reference `output`. Which one it sees first is fixed per
|
| 86 |
prompt, as in AlpacaEval. A sample's score is the judge's probability that the reply is
|
| 87 |
better, read from its top-5 logprobs.
|
| 88 |
-
- The
|
| 89 |
-
than 500 characters longer than its reference counts as a tie at
|
| 90 |
-
|
| 91 |
|
| 92 |
## Construction
|
| 93 |
|
|
@@ -121,8 +115,5 @@ pins this dataset's `gold` config to a fixed revision.
|
|
| 121 |
```sh
|
| 122 |
hf download HuggingFaceSwarms/SmolTalkBench scripts/create_dataset.py --repo-type dataset --local-dir .
|
| 123 |
uv run scripts/create_dataset.py --n 1000 --seed 0 \
|
| 124 |
-
--source_revision 5feaf2fd3ffca7c237fc38d1861bc30365d48ffa --repo HuggingFaceSwarms/SmolTalkBench --push
|
| 125 |
```
|
| 126 |
-
|
| 127 |
-
The chart in `assets/progress.png` is not built by this script: it plots the PyBot Space's
|
| 128 |
-
progress data, which comes from the leaderboard's results.
|
|
|
|
| 61 |
|
| 62 |
## What it's for
|
| 63 |
|
| 64 |
+
SmolTalkBench was built to measure [PyBot](https://huggingface.co/spaces/HuggingFaceSwarms/PyBot),
|
| 65 |
+
a chatbot that a swarm of agents wrote entirely in Python: no weights, no training, just
|
| 66 |
+
if-else branches, regex patterns and lookup tables. Every version of PyBot was scored on
|
| 67 |
+
SmolTalkBench, with small instruction-tuned LMs as reference points. Its win rate climbed from
|
| 68 |
+
0% to 26% over more than 6,000 commits, on par with SmolLM-360M-Instruct.
|
|
|
|
|
|
|
|
|
|
| 69 |
|
| 70 |

|
| 71 |
|
|
|
|
| 74 |
|
| 75 |
## How it's scored
|
| 76 |
|
|
|
|
|
|
|
|
|
|
| 77 |
- The model answers each `instruction` once, as a single user turn.
|
| 78 |
- An LLM judge (`deepseek-ai/DeepSeek-V4.1-Flash`, with AlpacaEval 2.0's weighted annotator
|
| 79 |
prompt) compares the reply with the reference `output`. Which one it sees first is fixed per
|
| 80 |
prompt, as in AlpacaEval. A sample's score is the judge's probability that the reply is
|
| 81 |
better, read from its top-5 logprobs.
|
| 82 |
+
- The win rate is the mean score with [WildBench](https://github.com/allenai/WildBench)'s
|
| 83 |
+
length penalty: a reply more than 500 characters longer than its reference counts as a tie at
|
| 84 |
+
most. PyBot's win rates above use it.
|
| 85 |
|
| 86 |
## Construction
|
| 87 |
|
|
|
|
| 115 |
```sh
|
| 116 |
hf download HuggingFaceSwarms/SmolTalkBench scripts/create_dataset.py --repo-type dataset --local-dir .
|
| 117 |
uv run scripts/create_dataset.py --n 1000 --seed 0 \
|
| 118 |
+
--source_revision 5feaf2fd3ffca7c237fc38d1861bc30365d48ffa --repo HuggingFaceSwarms/SmolTalkBench --push --public
|
| 119 |
```
|
|
|
|
|
|
|
|
|
scripts/create_dataset.py
CHANGED
|
@@ -97,14 +97,11 @@ instructions so that results compare across them.
|
|
| 97 |
|
| 98 |
## What it's for
|
| 99 |
|
| 100 |
-
SmolTalkBench
|
| 101 |
-
|
| 102 |
-
|
| 103 |
-
|
| 104 |
-
|
| 105 |
-
LMs, against the same fixed references. The best of these bots is
|
| 106 |
-
[PyBot](https://github.com/huggingface/pybot): its win rate climbed from 0% to 26% over more
|
| 107 |
-
than 6,000 commits, on par with SmolLM-360M-Instruct.
|
| 108 |
|
| 109 |

|
| 110 |
|
|
@@ -113,17 +110,14 @@ baselines' scores are inflated.
|
|
| 113 |
|
| 114 |
## How it's scored
|
| 115 |
|
| 116 |
-
The eval is the Inspect task `smoltalk_bench_gold` in flock's `experiments/run4/eval`, which
|
| 117 |
-
pins this dataset's `{CONFIG}` config to a fixed revision.
|
| 118 |
-
|
| 119 |
- The model answers each `instruction` once, as a single user turn.
|
| 120 |
- An LLM judge (`deepseek-ai/DeepSeek-V4.1-Flash`, with AlpacaEval 2.0's weighted annotator
|
| 121 |
prompt) compares the reply with the reference `output`. Which one it sees first is fixed per
|
| 122 |
prompt, as in AlpacaEval. A sample's score is the judge's probability that the reply is
|
| 123 |
better, read from its top-5 logprobs.
|
| 124 |
-
- The
|
| 125 |
-
than 500 characters longer than its reference counts as a tie at
|
| 126 |
-
|
| 127 |
|
| 128 |
## Construction
|
| 129 |
|
|
@@ -159,9 +153,6 @@ hf download {args.repo} {SCRIPT} --repo-type dataset --local-dir .
|
|
| 159 |
uv run {SCRIPT} --n {args.n} --seed {args.seed} \\
|
| 160 |
--source_revision {args.source_revision} --repo {args.repo} --push{public}
|
| 161 |
```
|
| 162 |
-
|
| 163 |
-
The chart in `assets/progress.png` is not built by this script: it plots the PyBot Space's
|
| 164 |
-
progress data, which comes from the leaderboard's results.
|
| 165 |
"""
|
| 166 |
|
| 167 |
|
|
@@ -195,6 +186,7 @@ def main():
|
|
| 195 |
if entry["config_name"] == CONFIG:
|
| 196 |
entry["default"] = True
|
| 197 |
pushed.text = card(args, len(rows))
|
|
|
|
| 198 |
api = HfApi()
|
| 199 |
api.create_commit(args.repo, repo_type="dataset", operations=[
|
| 200 |
CommitOperationAdd("README.md", str(pushed).encode()),
|
|
|
|
| 97 |
|
| 98 |
## What it's for
|
| 99 |
|
| 100 |
+
SmolTalkBench was built to measure [PyBot](https://huggingface.co/spaces/HuggingFaceSwarms/PyBot),
|
| 101 |
+
a chatbot that a swarm of agents wrote entirely in Python: no weights, no training, just
|
| 102 |
+
if-else branches, regex patterns and lookup tables. Every version of PyBot was scored on
|
| 103 |
+
SmolTalkBench, with small instruction-tuned LMs as reference points. Its win rate climbed from
|
| 104 |
+
0% to 26% over more than 6,000 commits, on par with SmolLM-360M-Instruct.
|
|
|
|
|
|
|
|
|
|
| 105 |
|
| 106 |

|
| 107 |
|
|
|
|
| 110 |
|
| 111 |
## How it's scored
|
| 112 |
|
|
|
|
|
|
|
|
|
|
| 113 |
- The model answers each `instruction` once, as a single user turn.
|
| 114 |
- An LLM judge (`deepseek-ai/DeepSeek-V4.1-Flash`, with AlpacaEval 2.0's weighted annotator
|
| 115 |
prompt) compares the reply with the reference `output`. Which one it sees first is fixed per
|
| 116 |
prompt, as in AlpacaEval. A sample's score is the judge's probability that the reply is
|
| 117 |
better, read from its top-5 logprobs.
|
| 118 |
+
- The win rate is the mean score with [WildBench](https://github.com/allenai/WildBench)'s
|
| 119 |
+
length penalty: a reply more than 500 characters longer than its reference counts as a tie at
|
| 120 |
+
most. PyBot's win rates above use it.
|
| 121 |
|
| 122 |
## Construction
|
| 123 |
|
|
|
|
| 153 |
uv run {SCRIPT} --n {args.n} --seed {args.seed} \\
|
| 154 |
--source_revision {args.source_revision} --repo {args.repo} --push{public}
|
| 155 |
```
|
|
|
|
|
|
|
|
|
|
| 156 |
"""
|
| 157 |
|
| 158 |
|
|
|
|
| 186 |
if entry["config_name"] == CONFIG:
|
| 187 |
entry["default"] = True
|
| 188 |
pushed.text = card(args, len(rows))
|
| 189 |
+
# The card's chart, assets/progress.png, is uploaded separately and left in place here.
|
| 190 |
api = HfApi()
|
| 191 |
api.create_commit(args.repo, repo_type="dataset", operations=[
|
| 192 |
CommitOperationAdd("README.md", str(pushed).encode()),
|