Update leaderboard, and documentation.

benchmark.csv

@@ -1,7 +1,7 @@
-model,ivrit-ai/eval-d1,ivrit-ai/saspeech,google/fleurs/he,mozilla-foundation/common_voice_17_0/he
+model,ivrit-ai/eval-d1,ivrit-ai/saspeech,google/fleurs/he,mozilla-foundation/common_voice_17_0/he,imvladikon/hebrew_speech_kan
 ivrit-ai/whisper-large-v2-d4,0.063,0.080,0.242,0.207
 ivrit-ai/whisper-v2-d3-e3,0.069,0.086,0.256,0.214
-openai/whisper-large-v2,0.083,0.100,0.276,0.234
+openai/whisper-large-v2,0.083,0.100,0.276,0.234,0.209
 openai/whisper-large-v3,0.101,0.096,0.262,0.232
-aws-transcribe-batch-20241205,0.069,0.087,0.230,0.141
-aws-transcribe-stream-20241205,0.082,0.092,0.287,0.199
+aws-transcribe-batch-20241205,0.069,0.087,0.230,0.141,0.092
+aws-transcribe-stream-20241205,0.082,0.092,0.287,0.199,0.132

src/about.py

@@ -21,50 +21,112 @@ NUM_FEWSHOT = 0 # Change with your few shot
 
 
 # Your leaderboard name
-TITLE = """<h1 align="center" id="space-title">Demo leaderboard</h1>"""
+TITLE = """<h1 align="center" id="space-title">Hebrew Speech Recognition Leaderboard</h1>"""
 
 # What does your leaderboard evaluate?
 INTRODUCTION_TEXT = """
-Intro text
+Welcome to the Hebrew Speech Recognition Leaderboard! This is a community-driven effort to track and compare the performance 
+of various speech recognition models on Hebrew language tasks.
+
+This leaderboard is maintained by [ivrit.ai](https://ivrit.ai), a project dedicated to advancing Hebrew language AI technologies. 
+You can find our work on [GitHub](https://github.com/ivrit-ai) and [Hugging Face](https://huggingface.co/ivrit-ai).
+
+## Motivation
+Hebrew presents unique challenges for speech recognition due to its rich morphology, absence of written vowels, and diverse 
+dialectal variations. This leaderboard aims to:
+- Provide standardized benchmarks for Hebrew ASR evaluation
+- Track progress in Hebrew speech recognition technology
+- Foster collaboration in the Hebrew NLP community
+- Make Hebrew speech technology more accessible
+
+## Benchmarks
+The following datasets are used in our evaluation:
+
+### [ivrit-ai/eval-d1](https://huggingface.co/datasets/ivrit-ai/eval-d1)
+- **Size**: 2 hours
+- **Domain**: Manual transcription of podcasts. Typical segment length is 5 minutes.
+- **Source**: Description of source
+
+### [ivrit-ai/saspeech](https://huggingface.co/datasets/ivrit-ai/saspeech)
+- **Size**: X hours
+- **Domain**: Description
+- **Source**: Description of source
+
+### [google/fleurs/he](https://huggingface.co/datasets/google/fleurs)
+- **Size**: X hours
+- **Domain**: Description
+- **Source**: Description of source
+
+### [mozilla-foundation/common_voice_17_0/he](https://huggingface.co/datasets/mozilla-foundation/common_voice_17_0)
+- **Size**: X hours
+- **Domain**: Description
+- **Source**: Description of source
+
+### [imvladikon/hebrew_speech_kan](https://huggingface.co/datasets/imvladikon/hebrew_speech_kan)
+- **Size**: X hours
+- **Domain**: Description
+- **Source**: Description of source
 """
 
-# Which evaluations are you running? how can people reproduce what you have?
-LLM_BENCHMARKS_TEXT = f"""
+# Technical details about evaluation
+LLM_BENCHMARKS_TEXT = """
 ## How it works
+Models are evaluated using Word Error Rate (WER) on each benchmark dataset. The final score is an average of WER across all benchmarks, 
+with lower scores indicating better performance.
+
+Specifically, evaluation is done using the [jiwer](https://github.com/jitsi/jiwer) library.
+Source code for the evaluation can be found [here](https://github.com/ivrit-ai/asr-training/blob/master/evaluate_model.py).
 
 ## Reproducibility
-To reproduce our results, here is the commands you can run:
+To evaluate your model on these benchmarks, you can use our evaluation script as follows:
 
-"""
+```bash
+./evaluate_model.py --engine <engine> --model <model> --dataset <dataset:split:column> [--name <name>] [--workers <num_workers>]
+```
 
-EVALUATION_QUEUE_TEXT = """
-## Some good practices before submitting a model
+For example, here's how to evaluate ivrit-ai/faster-whisper-v2-d4 on the google/fleurs/he dataset:
 
-### 1) Make sure you can load your model and tokenizer using AutoClasses:
-```python
-from transformers import AutoConfig, AutoModel, AutoTokenizer
-config = AutoConfig.from_pretrained("your model name", revision=revision)
-model = AutoModel.from_pretrained("your model name", revision=revision)
-tokenizer = AutoTokenizer.from_pretrained("your model name", revision=revision)
+```bash
+./evaluate_model.py --engine faster-whisper --model ivrit-ai/faster-whisper-v2-d4 --name he_il --dataset google/fleurs:test:transcription --workers 1
 ```
-If this step fails, follow the error messages to debug your model before submitting it. It's likely your model has been improperly uploaded.
 
-Note: make sure your model is public!
-Note: if your model needs `use_remote_code=True`, we do not support this option yet but we are working on adding it, stay posted!
+"""
 
-### 2) Convert your model weights to [safetensors](https://huggingface.co/docs/safetensors/index)
-It's a new format for storing weights which is safer and faster to load and use. It will also allow us to add the number of parameters of your model to the `Extended Viewer`!
+EVALUATION_QUEUE_TEXT = """
+## Submitting a model for evaluation
 
-### 3) Make sure your model has an open license!
-This is a leaderboard for Open LLMs, and we'd love for as many people as possible to know they can use your model 🤗
+### 1) Provide an inference script
+To evaluate your model, we need either:
 
-### 4) Fill up your model card
-When we add extra information about models to the leaderboard, it will be automatically taken from the model card
+a) A simple inference script that takes audio input and returns transcribed text:
+```python
+def transcribe(audio_path: str) -> str:
+    # Your model loading and inference code here
+    return transcribed_text
+```
 
-## In case of model failure
-If your model is displayed in the `FAILED` category, its execution stopped.
-Make sure you have followed the above steps first.
-If everything is done, check you can launch the EleutherAIHarness on your model locally, using the above command without modifications (you can add `--limit` to limit the number of examples per task).
+b) Or augment our evaluate_model.py script with your model's implementation.
+
+### 2) Make sure your model is publicly accessible
+Your model should be available on the Hugging Face Hub with:
+- Public visibility
+- Clear licensing information
+- Basic model card documentation
+
+### 3) Fill up your model card
+Please include in your model card:
+- Model architecture
+- Training data description
+- Licensing information
+- Any special preprocessing requirements
+- Expected input format (sampling rate, audio format, etc.)
+
+## In case of evaluation failure
+If your model evaluation fails, please:
+1. Check that your model can be loaded and run locally
+2. Verify your inference script works with our benchmark format
+3. Ensure all dependencies are clearly specified
+4. Contact us through GitHub issues if problems persist
 """
 
 CITATION_BUTTON_LABEL = "Copy the following snippet to cite these results"