Initial BTZSC Phase 1 results dataset

README.md

@@ -1,6 +1,6 @@
 ---
 pretty_name: BTZSC Leaderboard Results
-license: apache-2.0
+license: mit
 tags:
   - leaderboard
   - text-classification
@@ -12,14 +12,21 @@ language:
 
 # BTZSC Results
 
-This repository stores model submissions for the **BTZSC** leaderboard:
+This repository stores model submissions for the **BTZSC** leaderboard.
 
 **BTZSC: A Benchmark for Zero-Shot Text Classification across Cross-Encoders, Embedding Models, Rerankers and LLMs**.
 
-- Paper: https://openreview.net/pdf?id=IxMryAz2p3
-- Eval harness: https://github.com/btzsc/btzsc
+- Paper: https://openreview.net/forum?id=IxMryAz2p3
+- Eval harness: https://github.com/IliasAarab/btzsc
 - Leaderboard Space: https://huggingface.co/spaces/btzsc/btzsc-leaderboard
 
+Benchmark summary:
+
+- 22 English single-label datasets
+- 4 task families: sentiment, topic, intent, emotion
+- Strict zero-shot protocol (no BTZSC-label training/tuning)
+- Primary metric: macro-F1
+
 ## What this repo contains
 
 - One JSON file per model evaluation run in `results/<model_type>/<model-name>.json`
@@ -34,10 +41,26 @@ Each submission follows schema version `1.0` with:
 - `evaluation`: harness versioning and runtime metadata
 - `results.overall`: averaged macro F1 / accuracy / macro precision / macro recall
 - `results.by_task`: sentiment/topic/intent/emotion aggregates
-- `results.by_dataset`: per-dataset metric blocks (ground truth)
+- `results.by_dataset`: per-dataset metric blocks
 
 ## Contributing results
 
-See [SUBMISSION.md](SUBMISSION.md) for exact instructions.
+Destination path format:
+
+- `results/<model_type>/<model-name>.json`
+
+Recommended flow:
+
+1. Export with the official harness (`btzsc evaluate ... --output-json ...`).
+2. Validate locally (`python validate.py results/<model_type>/<model-name>.json`).
+3. Add your file at the required path.
+4. Submit by one of these methods:
+  - Web UI upload on Hugging Face (no clone required)
+  - Git workflow (direct push if you have write access, otherwise fork + PR)
+  - API workflow via `huggingface_hub` with `create_pr=True` (PR-based)
+
+In short: **add** means placing the JSON at the correct path; **submit** means publishing that change to this remote repo.
+
+See [SUBMISSION.md](SUBMISSION.md) for full requirements and review checks.
 
 PRs adding result files are validated in CI with `validate.py`.

SUBMISSION.md

@@ -40,12 +40,37 @@ python validate.py results/<model_type>/<model-name>.json
 
 ### 4) Open a Pull Request
 
-1. Fork this repository.
-2. Place your JSON file in `results/<model_type>/<model-name>.json`.
-3. Open a PR with:
-   - The result JSON file.
-   - A short model description (architecture/training notes).
-   - Confirmation that model weights are public on Hugging Face Hub.
+All submission actions in this step target the results dataset repo `btzsc/btzsc-results`: https://huggingface.co/datasets/btzsc/btzsc-results.
+
+Required destination path:
+
+- `results/<model_type>/<model-name>.json`
+
+Example:
+
+- `results/embedding/e5-base-v2.json`
+
+Choose one submission workflow:
+
+1. Web UI (no clone required)
+  - Open https://huggingface.co/datasets/btzsc/btzsc-results and upload the JSON in **Files and versions** at the required path.
+  - If you do not have write access, fork `btzsc/btzsc-results` and open a PR to `btzsc/btzsc-results`.
+
+2. Git workflow (clone/fork + push)
+  - Clone or fork `https://huggingface.co/datasets/btzsc/btzsc-results`.
+  - Add your JSON at the required path.
+  - Push directly (if you have write access) or push to your fork and open a PR to `btzsc/btzsc-results`.
+
+3. API workflow (`huggingface_hub`, PR-based)
+  - Authenticate first (`huggingface-cli login` or `HF_TOKEN`).
+  - Use `create_pr=True` against repo_id `btzsc/btzsc-results` to open a PR branch programmatically.
+  - If PR creation is restricted for your account, upload to your fork and open a PR to `btzsc/btzsc-results` manually.
+
+For every PR, include:
+
+- The result JSON file.
+- A short model description (architecture/training notes).
+- Confirmation that model weights are public on Hugging Face Hub.
 
 ## Merge checks