Upload processed TeleLogs dataset with MCQ format

INSTRUCTIONS.md

@@ -0,0 +1,93 @@
+# TeleLogs Dataset Extraction Instructions
+
+## Problem
+The current environment has network restrictions that block access to HuggingFace. The dataset cannot be downloaded automatically from `netop/TeleLogs`.
+
+## Solutions
+
+### Option 1: Run Locally (Recommended)
+
+Run the extraction script on your local machine where you have internet access to HuggingFace:
+
+```bash
+# Make sure you're in the project root
+cd /path/to/Telecom-Bench
+
+# Install dependencies
+pip install -e .
+
+# Set your HuggingFace token as an environment variable
+export HF_TOKEN=your_huggingface_token_here
+
+# Run the extraction script
+python extract_telelogs.py
+```
+
+The script will:
+1. Download the TeleLogs dataset from HuggingFace (using the provided token)
+2. Process the test split
+3. Transform the data:
+   - Remove 'C' prefix from answers (C4 → 4)
+   - Extract 8 choices into an array
+   - Clean the questions by removing template text
+4. Save to `telelogs/` folder in multiple formats (CSV, JSON, Parquet)
+
+### Option 2: Manual Download and Transform
+
+1. Download the dataset manually from https://huggingface.co/datasets/netop/TeleLogs
+2. Place the files in this `telelogs` folder
+3. Run a simplified transformation script (to be created if needed)
+
+### Option 3: Use Alternative Environment
+
+Run this in an environment without proxy restrictions:
+- Google Colab
+- Local Jupyter notebook
+- Any machine with direct internet access
+
+## Dataset Transformation Details
+
+The script performs these transformations on the test split:
+
+1. **Answer column**: Strips 'C' prefix and converts to 0-based index
+   - Before: `C1` -> After: `0`
+   - Before: `C4` -> After: `3`
+   - Before: `C8` -> After: `7`
+   - This matches the array indexing of the choices
+
+2. **Choices column** (new): Extracts 8 choices cleanly from question text
+   - Parses choices formatted as `C1: content`, `C2: content`, ..., `C8: content`
+   - For C8, stops at `\n\n` which separates the choice from actual question data
+   - Creates array: `["choice1", "choice2", ..., "choice8"]`
+   - Choices are indexed 0-7 in the array
+
+3. **Question column**: Extracts actual question data only
+   - **Removes** the instruction template completely (e.g., "Analyze the 5G wireless network...")
+   - **Removes** the choice listing (C1-C8)
+   - **Keeps only** the actual question data after `\n\n` following C8
+   - Typically starts with "Given:" and includes all data tables
+
+## Output Files
+
+After successful extraction, you'll have:
+- `telelogs_test.csv` - CSV format
+- `telelogs_test.json` - JSON format
+- `telelogs_test.parquet` - Parquet format (most efficient)
+- `README.md` - Dataset documentation
+
+## HuggingFace Authentication
+
+You'll need a HuggingFace token with access to the `netop/TeleLogs` dataset. Set it as an environment variable:
+
+```bash
+export HF_TOKEN=your_huggingface_token_here
+```
+
+Alternatively, you can use `huggingface-cli login` if you have the HuggingFace CLI installed.
+
+## Next Steps
+
+Once you've extracted the dataset:
+1. Commit the processed files to the repository
+2. The data will be available for the LLM evaluation pipeline
+3. You can work with the dataset in MCQ format

README.md

@@ -0,0 +1,20 @@
+# TeleLogs Dataset (Processed)
+
+This dataset has been extracted from HuggingFace and processed into MCQ format.
+
+## Transformations Applied
+
+1. **Answer column**: Removed 'C' prefix, keeping only the number (C4 -> 4)
+2. **Choices column**: Extracted 8 choices from questions into an array
+3. **Question column**: Removed template text and choice options
+
+## Dataset Info
+
+- Number of samples: 864
+- Columns: question, answer, choices
+
+## Files
+
+- `telelogs_test.csv`: CSV format
+- `telelogs_test.json`: JSON format
+- `telelogs_test.parquet`: Parquet format (recommended for large datasets)

telelogs_test.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:04aaa5e54662c62dfd756613cf03f7af4fb5a23bd660a9b73a8c06e0641626dd
+size 390128