Update dataset card with size info, data fields table, list-files example

README.md

@@ -35,7 +35,8 @@ datasets:
 
 | Attribute | Value |
 |-----------|-------|
-| Files | **16** |
+| Files | **16** (15 files + 1 subdirectory) |
+| Total size | **~141 KB** |
 | Created | 2026-07-06 |
 | Last updated | 2026-07-30 |
 | Downloads | **184** (as of last check) |
@@ -43,48 +44,55 @@ datasets:
 
 ---
 
-## Data Fields — Files & Purpose
+## Data Fields — All Files
 
-This dataset is a **collection of Python scripts and Jupyter notebooks**, not a tabular dataset. Each file serves a specific role in the SakThai model lifecycle:
+This dataset is a **collection of Python scripts and Jupyter notebooks**, not a tabular dataset. Each file serves a specific role in the SakThai model lifecycle. Below is the complete file listing:
 
 ### Training Notebooks
 
-| File | Purpose |
-|------|---------|
-| `sakthai-engine.ipynb` | Training notebook for the 1.5B model (Qwen2.5-1.5B-Instruct base, LoRA on combined-v6) |
-| `sakthai-7b-engine.ipynb` | Training notebook for the 7B model (Qwen2.5-7B-Instruct base, LoRA on combined-v6) |
+| File | Size | Purpose |
+|------|------|---------|
+| `sakthai-engine.ipynb` | 10.5 KB | Training notebook for the 1.5B model (Qwen2.5-1.5B-Instruct base, LoRA on combined-v6) |
+| `sakthai-7b-engine.ipynb` | 24.0 KB | Training notebook for the 7B model (Qwen2.5-7B-Instruct base, LoRA on combined-v6) |
 
 ### Job Scripts (0.5B)
 
-| File | Purpose |
-|------|---------|
-| `train-sakthai-0.5b-v2.py` | Improved 0.5B fine-tune (MLP LoRA targets, r16/rsLoRA, completion-only loss, v6 + irrelevance) |
-| `job-0.5b-v7.py` | 0.5B config-upgrade fine-tune on **combined-v7**, scored on sakthai-bench-v1. Uses `report_to="none"` and `save_strategy="no"` to avoid Trackio serialisation crashes |
-| `job-0.5b-hfjobs.py` | 0.5B config-upgrade fine-tune with manual ChatML render + BFCL before/after eval. Designed for HF Jobs |
-| `job-0.5b-exp.py` | 0.5B improvement experiments. Two modes via `SAK_MODE`: **lora-masked** (LoRA + prompt masking) and **full-masked** (full fine-tune + prompt masking) |
-| `job-0.5b-nanguard.py` | Guard experiments — isolation of prompt masking vs full fine-tune (same structure as `job-0.5b-exp.py`) |
-| `job-0.5b-nanhunt.py` | NanHunt experiments — further ablation studies on prompt masking variants |
+| File | Size | Purpose |
+|------|------|---------|
+| `train-sakthai-0.5b-v2.py` | 6.8 KB | Improved 0.5B fine-tune (MLP LoRA targets, r16/rsLoRA, completion-only loss, v6 + irrelevance) |
+| `job-0.5b-v7.py` | 10.2 KB | 0.5B config-upgrade fine-tune on **combined-v7**, scored on sakthai-bench-v1. Uses `report_to="none"` and `save_strategy="no"` to avoid Trackio serialisation crashes |
+| `job-0.5b-hfjobs.py` | 7.8 KB | 0.5B config-upgrade fine-tune with manual ChatML render + BFCL before/after eval. Designed for HF Jobs |
+| `job-0.5b-exp.py` | 12.9 KB | 0.5B improvement experiments. Two modes via `SAK_MODE`: **lora-masked** (LoRA + prompt masking) and **full-masked** (full fine-tune + prompt masking) |
+| `job-0.5b-nanguard.py` | 13.2 KB | Guard experiments — isolation of prompt masking vs full fine-tune (same structure as `job-0.5b-exp.py`) |
+| `job-0.5b-nanhunt.py` | 11.8 KB | NanHunt experiments — further ablation studies on prompt masking variants |
 
 ### Training Scripts (1.5B, 7B)
 
-| File | Purpose |
-|------|---------|
-| `scripts/train-sakthai-1.5b-v2.py` | 1.5B v2 training script — improved LoRA config and data pipeline |
-| `scripts/sakthai-7b-post-train.py` | LoRA merge → GGUF convert → upload to HF Hub |
+| File | Size | Purpose |
+|------|------|---------|
+| `scripts/train-sakthai-1.5b-v2.py` | 4.5 KB | 1.5B v2 training script — improved LoRA config and data pipeline |
+| `scripts/sakthai-7b-post-train.py` | 13.7 KB | LoRA merge → GGUF convert → upload to HF Hub |
 
 ### Evaluation & Validation
 
-| File | Purpose |
-|------|---------|
-| `eval-bfcl-0.5b.py` | BFCL-style eval on the held-out `test` split (simple / parallel / irrelevance) |
-| `validate.py` | Validate every non-GPU part of `job-0.5b-v7.py` locally: renderer sanity, bench-exclusion filter, oracle-pass scorer, token-length distribution vs MAX_LEN |
-| `validate_exp.py` | Validate prompt/completion explode locally: row counts, assistant markers, gold tool call recovery, no bench row leak into training |
+| File | Size | Purpose |
+|------|------|---------|
+| `eval-bfcl-0.5b.py` | 3.9 KB | BFCL-style eval on the held-out `test` split (simple / parallel / irrelevance) |
+| `validate.py` | 4.9 KB | Validate every non-GPU part of `job-0.5b-v7.py` locally: renderer sanity, bench-exclusion filter, oracle-pass scorer, token-length distribution vs MAX_LEN |
+| `validate_exp.py` | 3.7 KB | Validate prompt/completion explode locally: row counts, assistant markers, gold tool call recovery, no bench row leak into training |
 
 ### Deployment
 
-| File | Purpose |
-|------|---------|
-| `deploy-endpoint.py` | Deploy fine-tuned models to Hugging Face Inference Endpoints |
+| File | Size | Purpose |
+|------|------|---------|
+| `deploy-endpoint.py` | 3.1 KB | Deploy fine-tuned models to Hugging Face Inference Endpoints |
+
+### Repository Infrastructure
+
+| File | Size | Purpose |
+|------|------|---------|
+| `.gitattributes` | 2.4 KB | Git LFS configuration for dataset storage |
+| `README.md` | 7.0 KB | This dataset card — documentation and usage guide |
 
 ---
 
@@ -104,9 +112,11 @@ This dataset is a **collection of Python scripts and Jupyter notebooks**, not a
 
 3. **Validation scripts** — Two `validate*.py` scripts were added to prove every non-GPU component locally before launching paid jobs, catching data pipeline bugs early.
 
+4. **Experimental scripts** — `job-0.5b-exp.py`, `job-0.5b-nanguard.py`, and `job-0.5b-nanhunt.py` capture specific run configurations for reproducibility of ablation studies on prompt masking and LoRA strategies.
+
 **Filtering criteria.** Scripts were included if they:
 - Directly contribute to training, evaluating, validating, or deploying a SakThai model
-- Are self-contained (PEP 723 inline script metadata for dependencies)
+- Are self-contained (PEP 723 inline script metadata for dependencies where applicable)
 - Are licensed Apache 2.0
 - Are functionally distinct (duplicate-mode scripts like `nanguard`/`nanhunt` are included because they capture specific run configurations for reproducibility)
 
@@ -124,6 +134,18 @@ ds = load_dataset("Nanthasit/sakthai-kaggle-notebooks", split="train")
 print(ds)
 ```
 
+### List available files
+
+```python
+from datasets import get_dataset_split_names
+from huggingface_hub import HfApi
+
+api = HfApi()
+files = api.list_repo_files("Nanthasit/sakthai-kaggle-notebooks", repo_type="dataset")
+for f in sorted(files):
+    print(f)
+```
+
 ### Download individual scripts
 
 ```python