Clarify HF runner names and descriptions

README.md

@@ -1,5 +1,5 @@
 ---
-title: LBW Guard Colab Tests
+title: LBW Guard HF Evaluation Runner
 emoji: 🚀
 colorFrom: green
 colorTo: blue
@@ -23,17 +23,19 @@ Copyright (c) Qluon Inc. All rights reserved.
 
 Provided for Learn-By-Wire Guard evaluation and customer testing under the applicable Qluon license terms.
 
-# LBW Guard Colab Tests
+# LBW Guard HF Evaluation Runner
 
-This private Space runs notebook-faithful Hugging Face versions of:
+This private Hugging Face Space provides two GPU-ready LBW Guard evaluation runners:
 
-- `LBW_Guard_Easy_Test_COLAB.ipynb`
-- `LBW_Guard_Ablation_Test_COLAB.ipynb`
+- **Quick Comparison**: a short AdamW vs `lbw_guard` WikiText LoRA run based on `LBW_Guard_Easy_Test_COLAB.ipynb`.
+- **Ablation Matrix**: a scenario sweep for optimizer, learning rate, schedule, steps, data size, and LoRA rank based on `LBW_Guard_Ablation_Test_COLAB.ipynb`.
+
+Both runners produce JSON and CSV artifacts in the Space working directory. The quick runner writes final metrics plus LBW-vs-AdamW gains. The ablation runner writes per-scenario metrics plus LBW-vs-AdamW gains for each scenario.
 
 It installs `lbw-guard` from PyPI and does not vendor the local `lbw/` source folder.
 
 Paper: https://arxiv.org/abs/2605.19008
 
-Use GPU hardware for meaningful runtime. CPU can load the app, but training is intentionally capped to tiny smoke settings.
+Use GPU hardware for meaningful runtime. CPU can load the app, but training is intentionally capped to tiny smoke settings so users do not accidentally start long CPU jobs.
 
 The app writes run artifacts to the Space working directory. Add persistent storage if you need outputs to survive Space restarts.

app.py

@@ -521,7 +521,7 @@ def run_easy_test(
         ):
             yield (
                 "This Space is currently on `cpu-basic`. CPU mode is capped to 1 step, 20k train chars, "
-                "8k eval chars, and sampled validation. Switch the Space hardware to GPU for the Easy Test defaults.",
+                "8k eval chars, and sampled validation. Switch the Space hardware to GPU for the Quick Comparison defaults.",
                 None,
                 None,
                 None,
@@ -612,7 +612,8 @@ def run_easy_test(
 
         gains = _gain_rows(rows)
         payload = {
-            "source": "LBW_Guard_Easy_Test_COLAB.ipynb",
+            "source": "HF Quick Comparison Runner",
+            "based_on_colab": "LBW_Guard_Easy_Test_COLAB.ipynb",
             "config": {
                 "model_name": resolved_model,
                 "device": device_name,
@@ -644,9 +645,9 @@ def run_easy_test(
             "gains": gains,
             "logs": logs,
         }
-        json_path = run_dir / "lbw_guard_easy_test_results.json"
-        csv_path = run_dir / "lbw_guard_easy_test_results.csv"
-        gains_path = run_dir / "lbw_guard_easy_test_gains.csv"
+        json_path = run_dir / "lbw_guard_hf_quick_comparison_results.json"
+        csv_path = run_dir / "lbw_guard_hf_quick_comparison_results.csv"
+        gains_path = run_dir / "lbw_guard_hf_quick_comparison_gains.csv"
         json_path.write_text(json.dumps(payload, indent=2), encoding="utf-8")
         _write_csv(csv_path, rows)
         _write_csv(gains_path, gains)
@@ -1185,7 +1186,8 @@ def run_ablation_test(
 
         gains = _build_ablation_gain_rows(rows)
         payload = {
-            "source": "LBW_Guard_Ablation_Test_COLAB.ipynb",
+            "source": "HF Ablation Matrix Runner",
+            "based_on_colab": "LBW_Guard_Ablation_Test_COLAB.ipynb",
             "model_name": resolved_model,
             "device": device_name,
             "optimizers": optimizers,
@@ -1196,9 +1198,9 @@ def run_ablation_test(
             "gains": gains,
             "logs": logs,
         }
-        json_path = run_dir / "lbw_guard_ablation_results.json"
-        metrics_path = run_dir / "lbw_guard_ablation_metrics.csv"
-        gains_path = run_dir / "lbw_guard_ablation_gains.csv"
+        json_path = run_dir / "lbw_guard_hf_ablation_matrix_results.json"
+        metrics_path = run_dir / "lbw_guard_hf_ablation_matrix_metrics.csv"
+        gains_path = run_dir / "lbw_guard_hf_ablation_matrix_gains.csv"
         json_path.write_text(json.dumps(payload, indent=2), encoding="utf-8")
         _write_csv(metrics_path, rows)
         _write_csv(gains_path, gains)
@@ -1218,21 +1220,25 @@ def run_ablation_test(
 
 
 INTRO = """
-# LBW Guard Colab Tests
+# LBW Guard HF Evaluation Runner
 
-Runs notebook-faithful Hugging Face Space versions of:
+This private Space has two HF-specific runners based on the customer Colab notebooks:
 
-- `LBW_Guard_Easy_Test_COLAB.ipynb`
-- `LBW_Guard_Ablation_Test_COLAB.ipynb`
+- **Quick Comparison**: a short AdamW vs `lbw_guard` WikiText LoRA run.
+- **Ablation Matrix**: a selectable scenario sweep across optimizer, LR, schedule, steps, data, and LoRA rank.
 
-Current hardware is detected at run time. GPU is recommended for the default Easy Test.
+Each run writes JSON and CSV artifacts. GPU hardware is recommended; CPU mode is only for tiny smoke checks.
 """
 
 
-with gr.Blocks(title="LBW Guard Colab Tests") as demo:
+with gr.Blocks(title="LBW Guard HF Evaluation Runner") as demo:
     gr.Markdown(INTRO)
     with gr.Tabs():
-        with gr.Tab("Easy Test"):
+        with gr.Tab("Quick Comparison"):
+            gr.Markdown(
+                "Runs the HF version of the Easy Test flow: one shared WikiText slice, then AdamW and optional "
+                "`lbw_guard` LoRA training with final perplexity and gain CSV output."
+            )
             with gr.Row():
                 easy_model_name = gr.Textbox(value="TinyLlama/TinyLlama_v1.1", label="Model")
                 easy_run_lbw_guard = gr.Checkbox(value=True, label="Run LBW Guard comparison")
@@ -1252,7 +1258,7 @@ with gr.Blocks(title="LBW Guard Colab Tests") as demo:
                 easy_full_wikitext_train = gr.Checkbox(value=False, label="Full WikiText train")
                 easy_full_wikitext_eval = gr.Checkbox(value=False, label="Full WikiText eval")
                 easy_full_validation_ppl = gr.Checkbox(value=False, label="Full validation PPL")
-            easy_run_button = gr.Button("Run Easy Test", variant="primary")
+            easy_run_button = gr.Button("Run Quick Comparison", variant="primary")
             easy_summary = gr.Markdown()
             easy_json_file = gr.File(label="Raw JSON")
             easy_results_file = gr.File(label="Results CSV")
@@ -1279,7 +1285,11 @@ with gr.Blocks(title="LBW Guard Colab Tests") as demo:
                 outputs=[easy_summary, easy_json_file, easy_results_file, easy_gains_file],
             )
 
-        with gr.Tab("Ablation Test"):
+        with gr.Tab("Ablation Matrix"):
+            gr.Markdown(
+                "Runs the HF version of the ablation flow: build selected scenarios, run AdamW and optional "
+                "`lbw_guard` for each scenario, then export metrics and LBW-vs-AdamW gains."
+            )
             with gr.Row():
                 ablation_model_name = gr.Textbox(value="Qwen/Qwen2.5-0.5B", label="Model")
                 ablation_run_lbw_guard = gr.Checkbox(value=True, label="Run LBW Guard comparison")
@@ -1311,7 +1321,7 @@ with gr.Blocks(title="LBW Guard Colab Tests") as demo:
                 lr_sweep_text = gr.Textbox(value="1e-3, 5e-4", label="LR sweep")
                 step_sweep_text = gr.Textbox(value="100, 200", label="Step sweep")
                 lora_r_sweep_text = gr.Textbox(value="4, 8, 16", label="LoRA r sweep")
-            ablation_run_button = gr.Button("Run Ablation Test", variant="primary")
+            ablation_run_button = gr.Button("Run Ablation Matrix", variant="primary")
             ablation_summary = gr.Markdown()
             ablation_json_file = gr.File(label="Raw JSON")
             ablation_metrics_file = gr.File(label="Metrics CSV")