Use summary as metrics split name (all is reserved)

README.md

@@ -35,7 +35,7 @@ configs:
         path: predictions/outputs-*
   - config_name: metrics
     data_files:
-      - split: all
+      - split: summary
         path: predictions/metrics-*
   - config_name: glossary
     data_files:
@@ -110,7 +110,7 @@ Per-run aggregated metrics: one row per (model, configuration, language pair). T
 | `num_samples` | int | sentences scored |
 | `val_bleu` / `val_chrf` / `val_ssa_comet` | float | validation-split metrics where available |
 
-Single split: `all`.
+Single split: `summary`.
 
 ### `glossary`
 The co-developed bilingual scientific glossaries built during translation, one row per (English term, target-language translation) pair, stacked across all six target languages.
@@ -141,7 +141,7 @@ preds["outputs"][0]   # one model output per row
 
 # Per-run aggregated metrics.
 metrics = load_dataset("dsfsi/afriscience_mt", "metrics")
-metrics["all"][0]     # one (model, config, lang_pair) row
+metrics["summary"][0] # one (model, config, lang_pair) row
 
 # Bilingual scientific glossaries.
 gloss = load_dataset("dsfsi/afriscience_mt", "glossary")
@@ -155,7 +155,7 @@ Common joins:
 import pandas as pd
 test_corpus = corpus["test"].to_pandas()
 outputs     = preds["outputs"].to_pandas()
-metrics_df  = metrics["all"].to_pandas()
+metrics_df  = metrics["summary"].to_pandas()
 
 # Predictions paired with the source/reference from the corpus test split.
 joined = outputs.merge(