Upload README.md with huggingface_hub

README.md

@@ -1,13 +1,14 @@
 ---
 pretty_name: t2ranking-hard-neg-reasoning-embedding
-library_name: sentence-transformers
-pipeline_tag: sentence-similarity
 tags:
 - jsonl
 - retrieval
+- similarity
 - hard-negative
+- reasoning-embedding
 task_categories:
 - text-retrieval
+- sentence-similarity
 license: apache-2.0
 language:
 - multilingual
@@ -40,7 +41,7 @@ language:
 
 ## Introduction
 
-This is the dataset used to train the embedding models in the paper [`Do Reasoning Models Enhance Embedding Models?`](https://arxiv.org/abs/2601.21192). We use [`Qwen3-Embedding-0.6B`](https://huggingface.co/Qwen/Qwen3-Embedding-0.6B) to mine 3 hard negatives per query, and employ the positive-aware hard negative mining technique introduced in [`NV-Retriever`](https://arxiv.org/abs/2407.15831) with 95% margin to the positive score.
+This is the dataset used to train the embedding models in the paper [`Do Reasoning Models Enhance Embedding Models?`](https://arxiv.org/abs/2601.21192). We use [`Qwen3-Embedding-0.6B`](https://huggingface.co/Qwen/Qwen3-Embedding-0.6B) to mine 3 hard negatives per query, and employ the positive-aware hard negative mining technique introduced in [`NV-Retriever`](https://huggingface.co/nvidia/NV-Retriever-v1) with 95% margin to the positive score.
 
 ## Abstract
 
@@ -48,13 +49,13 @@ State-of-the-art embedding models are increasingly derived from decoder-only Lar
 
 ## Dataset Structure
 
-The dataset is structured according to the [`GritLM`](https://github.com/lucaswychan/gritlm-re) repository's format: `{"query": List[str], "pos": List[str], "neg": List[str]}`.
+The dataset is structured according to the [`GritLM`](https://github.com/lucaswychan/gritlm-re) repository's format: `{"query": List[str], "pos": List[str], "neg": List[str]}`. The script to mine the hard negatives is [here](https://github.com/HKUST-KnowComp/Reasoning-Embedding/blob/main/datasets/mine_hard_neg.py).
 
 *   **`query`**: This is a list containing two strings.
     *   `query[0]` holds the instruction. A complete list of instructions can be found [here](https://github.com/HKUST-KnowComp/Reasoning-Embedding/blob/main/evaluation/task_prompts.json).
     *   `query[1]` contains the actual query text.
 *   **`pos`**: A list with a single string, representing the positive anchor for the query. You can add more anchors to the list.
-*   **`neg`**: A list containing exactly three strings, which are the mined hard negatives associated with the query.
+*   **`neg`**: A list containing 1 - 3 strings, which are the mined hard negatives associated with the query.
 
 For example,