Add dataset card with provenance and licensing notes

README.md

@@ -1,42 +1,43 @@
 ---
-dataset_info:
-  features:
-  - name: messages
-    list:
-    - name: role
-      dtype: string
-    - name: content
-      dtype: string
-  - name: source
-    dtype: string
-  - name: source_dataset
-    dtype: string
-  - name: source_config
-    dtype: string
-  - name: source_split
-    dtype: string
-  - name: source_row
-    dtype: int64
-  - name: category
-    dtype: string
-  - name: utility
-    dtype: string
-  - name: variant_group
-    dtype: string
-  splits:
-  - name: train
-    num_bytes: 4110656991
-    num_examples: 1098277
-  - name: test
-    num_bytes: 214823943
-    num_examples: 54948
-  download_size: 2095293751
-  dataset_size: 4325480934
-configs:
-- config_name: default
-  data_files:
-  - split: train
-    path: data/train-*
-  - split: test
-    path: data/test-*
+license: other
+language:
+  - en
+task_categories:
+  - text-generation
+tags:
+  - conversational
+  - synthetic
+  - bash
+  - instruction-tuning
+  - provenance-preserved
+pretty_name: Bash Instruct III + SmolTalk
 ---
+
+# Bash Instruct III + SmolTalk
+
+This dataset combines the training split of [Bash Instruct III](https://huggingface.co/datasets/Frost2o24/bash-instruct-III-55k) with the `all` configuration of [SmolTalk](https://huggingface.co/datasets/HuggingFaceTB/smoltalk). SmolTalk's original test split remains the combined dataset's test split; Bash Instruct III is added to train only.
+
+All rows use chat-style `messages`. Provenance and source-specific fields are retained:
+
+- `source`: SmolTalk component name, or `bash-instruct-III-55k` for Bash rows.
+- `source_dataset`, `source_config`, `source_split`, `source_row`: exact upstream dataset and row location.
+- `category`, `utility`, `variant_group`: Bash Instruct III metadata; null for SmolTalk rows.
+
+## Source composition
+
+- Bash Instruct III: 54,360 training examples, licensed MIT. Its card also notes tldr-pages-derived seed phrasings under CC-BY-4.0.
+- SmolTalk `all`: the source's train and test splits, containing the documented SmolTalk subsets and several existing public datasets.
+
+## Licensing and attribution
+
+This combined dataset does not relicense its source material. SmolTalk documents Apache-2.0 for its newly created SmolTalk subsets and directs users to the original dataset licenses for included public datasets. Review those upstream terms, attributions, and any redistribution requirements before redistributing or using individual components. See the [SmolTalk dataset card](https://huggingface.co/datasets/HuggingFaceTB/smoltalk) and [Bash Instruct III dataset card](https://huggingface.co/datasets/Frost2o24/bash-instruct-III-55k).
+
+## Load
+
+```python
+from datasets import load_dataset
+
+dataset = load_dataset("OWNER/bash-smoltalk-combined")
+```
+
+The repository is intended to be private until access and redistribution terms for all SmolTalk components have been reviewed.