suchirsalhan's picture
Initial release: summer merge experiments (18,171 merges, 64,180 evaluations)
1e80dec verified
Raw History Blame Contribute Delete
5.87 kB
"""Render the Hugging Face dataset card (README.md) from the built tables. Never edited by hand."""
import pandas as pd
YAML = """---
license: cc-by-4.0
pretty_name: Model Merging & Capability Transfer Dataset
tags: [model-merging, permutation-alignment, mergeability, language-models, negative-results]
configs:
- config_name: merged_models
data_files: data/merged_models.parquet
default: true
- config_name: merges
data_files: data/merges.parquet
- config_name: evaluations
data_files: data/evaluations.parquet
- config_name: models
data_files: data/models.parquet
- config_name: summary
data_files: data/summary.parquet
---
"""
def _tbl(df):
cols = list(df.columns)
out = ["| " + " | ".join(map(str, cols)) + " |", "|" + "---|" * len(cols)]
for _, r in df.iterrows():
out.append("| " + " | ".join("" if pd.isna(v) else str(v) for v in r) + " |")
return "\n".join(out)
def render(merges, evals, models, flat, cov) -> str:
fam = merges.groupby("experiment_family").agg(
merges=("merge_id", "size"),
methods=("merge_method", lambda s: ", ".join(sorted(set(s)))),
unverified=("evaluation_status", lambda s: int((s != "complete").sum()))).reset_index()
ev = flat[flat["result_status"] == "evaluated"]
succ = ev.groupby("model_family").agg(
comparisons=("merge_success", "size"),
beats_best_parent=("merge_success", lambda s: int((s.astype(str) == "True").sum()))).reset_index()
status = flat["result_status"].value_counts().rename_axis("result_status").reset_index(name="rows")
return YAML + f"""
# Model Merging & Capability Transfer Dataset
A dataset of **merge experiments**, not just checkpoints. Every row says which two models were merged, how, under which alignment and configuration, and how the result scored on downstream tasks **next to both parents and the naive merge, measured under the same protocol**. Negative results, controls, unverified and incomplete runs are included, and their status is explicit.
It contains {cov['merges']:,} merge experiments, {cov['evaluations']:,} evaluations of {len(models):,} checkpoints, and {len(flat):,} rows in the flat `merged_models` table.
## Configurations
`merged_models` is the default configuration: one long-format row per pair × method × alignment × benchmark × task × metric, carrying parent, naive and merged scores and the derived merge loss, gain and success. `merges`, `evaluations` and `models` are the linked tables it is built from. `summary` holds the derived statistics.
```python
from datasets import load_dataset
flat = load_dataset("Cross-Mergeability/model-merging-benchmark", "merged_models")["train"]
evals = load_dataset("Cross-Mergeability/model-merging-benchmark", "evaluations")["train"]
```
## Experiment families
{_tbl(fam)}
## How often a merge beats its better parent
These counts are for comparisons where both parents and the merge were scored in the same experiment:
{_tbl(succ)}
## Metrics
All losses are signed so that **positive means worse than the better parent**:
- **Merge loss.** For NLL, `merged − min(A, B)`. For accuracy, `max(A, B) − merged`.
- **Merge gain.** `naive_merge_loss − method_merge_loss`. Positive means the method improved on plain averaging. It does **not** mean the result is useful.
- **Merge success.** The merge strictly beats the better parent on that metric.
- **Alignment gain.** Loss of the unaligned twin (same pair, method and configuration) minus loss of the aligned merge.
NLL and accuracy are never mixed in one row. A parent is used as a baseline only when it was scored in the same `experiment_id`.
## Row status (`result_status`)
{_tbl(status)}
- **`evaluated`:** the merge and both parents were scored under one protocol.
- **`parent_baseline_missing`:** the merge was scored but a parent was not, in that experiment. This happens, for example, for PolyPythia per-paradigm BLiMP, where only the parents' aggregate BLiMP was recorded, and for Beetle published NLL, where only the better-parent NLL was recorded.
- **`naive_baseline_missing`:** there is no plain-average twin in the same experiment.
- **`naive_baseline_not_applicable`:** the method has no plain-average twin by design (chat-vector transfer, OT fusion).
- **`unverified`:** the source flags the value.
Pair × method combinations that were never run have **no row**.
## Known limitations
- **Copies a parent.** `copies_parent=True` marks rungs whose output is one parent's tensors, for example SVC/OPCM single-parent fills. They are not evidence of composition.
- **Aligned equals naive.** In the published Beetle rungs, the aligned merges are byte-identical to the naive ones, because the aligner returned the identity; see `notes`.
- **Calibration overlap.**
- OT fusion is scored on sentences inside the set used to fit the transport plans, so it is `provisional`.
- PolyPythia alignment is fitted on the same FLORES blocks it is scored on.
- **Task arithmetic without a shared base.** Where the parents share no base, task arithmetic and TIES are parent interpolations, not conventional task vectors (see `merge_config`).
- **MergeBench.** Rows are published scores, not re-evaluated, and the 5-expert merges have no `parent_a`/`parent_b`.
- **Not recorded.** Parent checkpoint hashes, code commits, evaluation timestamps and most dataset revisions are absent from the historical result files and are left null.
- **Wikitext-scored transport rungs.** The 56 Beetle transport rungs scored on wikitext are `unverified`.
## Provenance and reproduction
Every row has a `source_file`, and `sources.json` lists every file each adapter read. Rebuild from the repository with `python -m merge_dataset.build_dataset`; the scripts are in `scripts/`. The tables are generated, never maintained by hand.
## Citation
To be added by the authors.
"""