Upload folder using huggingface_hub

README.md

@@ -4,253 +4,86 @@ task_categories:
   - text-classification
 language:
   - multilingual
-  - ar
-  - en
-  - fr
-  - es
-  - de
 tags:
   - language-identification
   - lid
   - benchmark
   - evaluation
-  - arabic-dialects
-  - multilingual
-pretty_name: LID Benchmark — Language Identification Evaluation Results
 size_categories:
-  - 10K<n<100K
+  - 10M<n<100M
 configs:
-  - config_name: results_per_language
-    data_files:
-      - split: train
-        path: data/results_per_language/train.parquet
-  - config_name: results_aggregate
-    data_files:
-      - split: train
-        path: data/results_aggregate/train.parquet
   - config_name: results_summary
-    data_files:
-      - split: train
-        path: data/results_summary/train.parquet
+    data_files: results_summary/train.parquet
+  - config_name: results_aggregate
+    data_files: results_aggregate/train.parquet
+  - config_name: results_per_language
+    data_files: results_per_language/train.parquet
+  - config_name: results_speed
+    data_files: results_speed/train.parquet
+  - config_name: model_languages
+    data_files: model_languages/train.parquet
+  - config_name: results_individual
+    data_files: results_individual/train.parquet
 ---
 
-# LID Benchmark — Language Identification Evaluation Results
-
-Structured evaluation results for **10 language identification models** across **8 benchmarks** covering **380 languages** — with per-language accuracy, aggregate metrics, and confusion analysis.
-
-Built as part of the [Gherbal](https://www.omneitylabs.com/models/gherbal) evaluation pipeline.
-
-The full [PDF report](https://www.omneitylabs.com/gherbal-v4/report.pdf) is also available.
-
-## Quick Start
-
-```python
-from datasets import load_dataset
-
-# Per-language results (26,540 rows)
-per_lang = load_dataset("omneity-labs/lid-benchmark", "results_per_language", split="train")
-
-# Aggregate metrics per model × benchmark × scope (400 rows)
-aggregate = load_dataset("omneity-labs/lid-benchmark", "results_aggregate", split="train")
-
-# Summary — one row per model × benchmark, full scope only (80 rows)
-summary = load_dataset("omneity-labs/lid-benchmark", "results_summary", split="train")
-```
-
-## Leaderboard (Full Scope)
-
-| Model | FLORES+ devtest | MADAR | Gherbal-Multi | ATLASIA-LID |
-|-------|:-:|:-:|:-:|:-:|
-| **GlotLID** | **0.9253** | 0.5648 | 0.7772 | 0.4977 |
-| OpenLID v2 | 0.8748 | 0.6262 | 0.7762 | 0.5735 |
-| OpenLID v3 (HPLT-LID) | 0.8556 | — | 0.6619 | — |
-| **Gherbal v4** | 0.8500 | **0.6298** | **0.8699** | **0.6909** |
-| OpenLID v1 | 0.8425 | 0.5587 | 0.8296 | 0.4845 |
-| NLLB-LID | 0.8331 | 0.1052 | 0.7522 | 0.3348 |
-| FastLID-176 | 0.4006 | 0.1352 | 0.6472 | 0.3899 |
-| Gherbal v3 | 0.3605 | 0.5745 | 0.8966 | 0.6561 |
-| Gherbal v2 | 0.1495 | 0.5811 | 0.7961 | 0.6561 |
-| Gherbal v1 | 0.1374 | 0.2771 | 0.8385 | 0.2718 |
-
-> **Note**: Full-scope FLORES+ accuracy penalizes models that support fewer languages (unsupported languages count as errors). Use `results_aggregate` with `scope=v4` (214 languages) for a fairer narrower comparison. Gherbal v4 achieves **0.9312** accuracy on FLORES+ devtest in the v4 scope.
-
-## Dataset Configs
-
-### `results_per_language` — Per-Language Breakdown
-
-26,540 rows. One row per (model, benchmark, scope, language).
-
-| Column | Type | Description |
-|--------|------|-------------|
-| `model` | string | Model name (e.g. `gherbal-v4`, `glotlid`) |
-| `benchmark` | string | Benchmark name (e.g. `flores-devtest`) |
-| `scope` | string | Language scope: `full`, `v1`, `v2`, `v3`, or `v4` |
-| `language` | string | Language code in `iso639-3_Script` format (e.g. `arb_Arab`) |
-| `n_samples` | int | Number of test samples for this language |
-| `accuracy` | float | Classification accuracy (0–1) |
-| `top_confusion_1` | string | Most confused-with language |
-| `top_confusion_1_count` | int | Count of samples misclassified as this language |
-| `top_confusion_2` | string | 2nd most confused-with language |
-| `top_confusion_2_count` | int | — |
-| `top_confusion_3` | string | 3rd most confused-with language |
-| `top_confusion_3_count` | int | — |
-| `confusions_json` | string | Full confusion map as JSON (all misclassified targets and counts) |
+# LID Benchmark
 
-**Example — find the hardest languages for a model:**
+Comprehensive evaluation of **17 language identification models** across **8 diverse benchmarks**.
 
-```python
-from datasets import load_dataset
-import pandas as pd
+Built by [Omneity Labs](https://www.omneitylabs.com).
 
-ds = load_dataset("omneity-labs/lid-benchmark", "results_per_language", split="train")
-df = ds.to_pandas()
+## Subsets
 
-# Worst-performing languages for Gherbal v4 on FLORES
-worst = (
-    df[(df["model"] == "gherbal-v4") &
-       (df["benchmark"] == "flores-devtest") &
-       (df["scope"] == "full") &
-       (df["n_samples"] >= 100)]
-    .sort_values("accuracy")
-    .head(10)
-    [["language", "accuracy", "n_samples", "top_confusion_1"]]
-)
-print(worst)
-```
+| Config | Description | Rows |
+|--------|-------------|------|
+| `results_summary` | One row per model × benchmark × scope with aggregate metrics | ~136 |
+| `results_aggregate` | Detailed aggregate metrics per model × benchmark × scope | ~816 |
+| `results_per_language` | Per-language accuracy for every model × benchmark × scope | ~57k |
+| `results_speed` | Inference speed (samples/sec) per model × benchmark | ~136 |
+| `model_languages` | Supported language codes declared by each model | ~4.7k |
+| `results_individual` | Every individual prediction (model × benchmark × sample) | ~28M |
 
-**Example — compare Arabic dialect accuracy across models:**
+## Models
 
-```python
-arabic_dialects = [
-    "arz_Arab", "ary_Arab", "arq_Arab", "aeb_Arab",
-    "apc_Arab", "acm_Arab", "ars_Arab", "afb_Arab", 
-    # Add more
-]
-
-arabic_df = df[
-    (df["language"].isin(arabic_dialects)) &
-    (df["benchmark"] == "flores-devtest") &
-    (df["scope"] == "full")
-]
-
-pivot = arabic_df.pivot_table(
-    index="language", columns="model", values="accuracy"
-)
-print(pivot.round(3))
-```
-
-### `results_aggregate` — Aggregate Metrics
-
-400 rows. One row per (model, benchmark, scope).
-
-| Column | Type | Description |
-|--------|------|-------------|
-| `model` | string | Model name |
-| `benchmark` | string | Benchmark name |
-| `scope` | string | Language scope |
-| `accuracy` | float | Overall accuracy |
-| `f1_macro` | float | Macro-averaged F1 |
-| `f1_weighted` | float | Weighted F1 |
-| `precision_macro` | float | Macro-averaged precision |
-| `recall_macro` | float | Macro-averaged recall |
-| `n_samples` | int | Total evaluation samples |
-| `n_classes` | int | Number of unique languages |
-
-**Example — model comparison across scopes:**
-
-```python
-ds = load_dataset("omneity-labs/lid-benchmark", "results_aggregate", split="train")
-df = ds.to_pandas()
-
-comparison = df[
-    (df["benchmark"] == "flores-devtest") &
-    (df["model"].isin(["gherbal-v4", "glotlid", "openlid-v2"]))
-].pivot_table(index="scope", columns="model", values="accuracy")
-print(comparison.round(4))
-```
-
-### `results_summary` — Quick Summary
-
-80 rows. One row per (model, benchmark) — full scope only. Best for quick leaderboard construction.
-
-| Column | Type | Description |
-|--------|------|-------------|
-| `model` | string | Model name |
-| `benchmark` | string | Benchmark name |
-| `accuracy` | float | Overall accuracy (full scope) |
-| `f1_macro` | float | Macro F1 (full scope) |
-| `f1_weighted` | float | Weighted F1 (full scope) |
-| `precision_macro` | float | Macro precision (full scope) |
-| `recall_macro` | float | Macro recall (full scope) |
-| `n_samples` | int | Total samples |
-| `n_classes` | int | Number of classes |
-
-## Models Evaluated
-
-| Model | Type | Languages | Source |
-|-------|------|-----------|--------|
-| [Gherbal v4](https://www.omneitylabs.com/models/gherbal) | FastText | 214 | Omneity Labs |
-| Gherbal v3 | FastText | 106 | Omneity Labs |
-| Gherbal v2 | FastText | 46 | Omneity Labs |
-| Gherbal v1 | FastText | 36 | Omneity Labs |
-| [GlotLID v3](https://huggingface.co/cis-lmu/glotlid) | FastText | 2,102 | LMU Munich |
-| [NLLB-LID](https://huggingface.co/facebook/fasttext-language-identification) | FastText | 218 | Meta |
-| [OpenLID v1](https://huggingface.co/laurievb/OpenLID) | FastText | 201 | Laurie Burchell |
-| [OpenLID v2](https://huggingface.co/laurievb/OpenLID) | FastText | 201 | Laurie Burchell |
-| [OpenLID v3 (HPLT-LID)](https://huggingface.co/HPLT/hplt_lid_fl) | FastText | 201 | HPLT |
-| [FastLID-176](https://fasttext.cc/docs/en/language-identification.html) | FastText | 176 | Meta |
+gherbal-v1, gherbal-v2, gherbal-v3, gherbal-v4, nllb-lid, openlid-v1, openlid-v2,
+hplt-openlid-v3, fastlid-176, glotlid, franc, franc-all, franc-min, cld2, langdetect,
+langid, py3langid
 
 ## Benchmarks
 
-| Benchmark | Samples | Languages | Description |
-|-----------|---------|-----------|-------------|
-| **FLORES+ devtest** | 222,640 | 214 | [openlanguagedata/flores_plus](https://huggingface.co/datasets/openlanguagedata/flores_plus) devtest split |
-| **FLORES+ dev** | 224,325 | 220 | [openlanguagedata/flores_plus](https://huggingface.co/datasets/openlanguagedata/flores_plus) dev split |
-| **MADAR** | 5,600 | 15 | [sawalni-ai/madar](https://huggingface.co/datasets/sawalni-ai/madar) — Arabic dialect corpus |
-| **Gherbal-Multi** | 185,000+ | 106+ | [sawalni-ai/gherbal-multi](https://huggingface.co/datasets/sawalni-ai/gherbal-multi) — multi-source test set |
-| **ATLASIA-LID** | 234,000+ | 24 | [atlasia/Arabic-LID-Leaderboard](https://huggingface.co/datasets/atlasia/Arabic-LID-Leaderboard) — Arabic country-level dialects |
-| **WiLI-2018** | — | 235 | Wikipedia Language Identification |
-| **CommonLID** | — | — | Common Crawl language ID |
-| **Bouquet** | — | — | Cross-domain evaluation |
-
-## Evaluation Scopes
-
-Results include multiple **scopes** to enable fair comparison between models with different language coverage:
+| Benchmark | Source |
+|-----------|--------|
+| flores-devtest | [openlanguagedata/flores_plus](https://huggingface.co/datasets/openlanguagedata/flores_plus) (devtest split) |
+| flores-dev | [openlanguagedata/flores_plus](https://huggingface.co/datasets/openlanguagedata/flores_plus) (dev split) |
+| madar | [Madar](https://camel.abudhabi.nyu.edu/madar-parallel-corpus) |
+| gherbal-multi | [sawalni-ai/gherbal-multi](https://huggingface.co/datasets/sawalni-ai/gherbal-multi) |
+| atlasia-lid | [atlasia/Arabic-LID-Leaderboard](https://huggingface.co/datasets/atlasia/Arabic-LID-Leaderboard) |
+| wili-2018 | [wili_2018](https://huggingface.co/datasets/wili_2018) |
+| commonlid | [commoncrawl/CommonLID](https://huggingface.co/datasets/commoncrawl/CommonLID) |
+| bouquet | [facebook/bouquet](https://huggingface.co/datasets/facebook/bouquet) |
 
-| Scope | Languages | Description |
-|-------|-----------|-------------|
-| `full` | All | All languages in the benchmark (penalizes models with fewer supported languages) |
-| `v1` | 36 | Intersection with Gherbal v1 language set |
-| `v2` | 46 | Intersection with Gherbal v2 language set |
-| `v3` | 106 | Intersection with Gherbal v3 language set |
-| `v4` | 214 | Intersection with Gherbal v4 language set |
+## Methodology
 
-Using scoped evaluation ensures models are compared only on languages they were designed to handle. For example, Gherbal v3 supports 106 languages — its `v3` scope accuracy on FLORES+ is much higher than its `full` scope accuracy, because the full scope includes 108+ languages it was never trained on.
+All predictions are normalized to **ISO 639-3 + Script** (ISO 15924) codes using [babelcode](https://github.com/omneity-labs/babelcode).
+Metrics: accuracy, macro-F1, weighted-F1, precision, recall — computed under multiple scopes (full, self, v1–v4).
 
-## Language Codes
+## Interactive App
 
-Languages use the `iso639-3_Script` format from FLORES+:
-- `arb_Arab` — Modern Standard Arabic (Arabic script)
-- `arz_Arab` — Egyptian Arabic
-- `ary_Arab` — Moroccan Arabic (Arabic script)
-- `ary_Latn` — Moroccan Arabic (Latin script)
-- `eng_Latn` — English
-- `fra_Latn` — French
-
-Full list of 380 languages available in the `results_per_language` config.
-
-## CSV Downloads
-
-For convenience, CSV versions of all three configs are also included in the `csv/` directory.
+Explore results interactively: [LID Benchmark Leaderboard](https://huggingface.co/spaces/omneity-labs/lid-benchmark)
 
 ## Citation
 
 If you use this benchmark data in your research, please reference:
 
-- **Omneity Labs LID Benchmark**: [https://huggingface.co/datasets/omneity-labs/lid-benchmark](https://huggingface.co/datasets/omneity-labs/lid-benchmarkk)
-- **Gherbal model**: [https://www.omneitylabs.com/models/gherbal](https://www.omneitylabs.com/models/gherbal)
-- **Evaluation benchmarks**: See individual benchmark datasets linked above.
+- Omneity Labs LID Benchmark: https://huggingface.co/datasets/omneity-labs/lid-benchmark
+- Gherbal model: https://www.omneitylabs.com/models/gherbal
+- Evaluation benchmarks: See individual benchmark datasets linked above.
+
+## Author
+
+- [Omar Kamali](https://omarkamali.com)
+- [Omneity Labs](https://omneitylabs.com)
 
 ## License
 
-The evaluation results in this dataset are released under [Apache 2.0](https://www.apache.org/licenses/LICENSE-2.0). The underlying benchmark datasets retain their original licenses.
+The evaluation results in this dataset are released under Apache 2.0. The underlying benchmark datasets retain their original licenses.

model_languages/train.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6876caff413c758d056de7fd6108f5db849020d3f3318d56d9bc53ac11c5c06d
+size 18754

results_aggregate/train.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3bf20ce860d68f7175f014b525107b5d7c58a6f6641483c69aaaf658e47b547d
+size 37711

results_individual/train.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:662733ab641f56ec4a95f8808e32ff3c11a586745e764510e8e96ec1400cd0d1
+size 531940611

results_per_language/train.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0258d7f2f8838317cac7f30e23a9decb7556e625b919b1be28357e5c26e1d6d8
+size 825967

results_speed/train.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:572ddf5f06d8687fccfb3e0eed159501b5f1bb62e037d13f0101457a74b3c693
+size 6416

results_summary/train.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f545437e1975df32822427e5c1876f33cc1ca21f048b54901cd1ff9ddda67a8b
+size 15447