Redact sensitive operational guidance from UI text and labels.

README.md

@@ -6,6 +6,7 @@ python_version: "3.10"
 app_file: app.py
 pinned: false
 emoji: 🧮
+
 ---
 
 # Math Conjecture Trainer Space
@@ -26,17 +27,9 @@ This Space is the tactical operations console for `maths-conjuncture-solutions`
 5. Apply quality gate thresholds before hub push.
 6. Emit `training_summary.json` + `post_eval_report.json` and stream live telemetry in UI.
 
-## Autonomous authentication
-
-No token input is required in the UI.
-
-Resolution order:
-
-1. `HF_TOKEN`
-2. `HUGGINGFACE_HUB_TOKEN`
-3. `huggingface-api-key.json`
+## Access posture
 
-If no token is available, public dataset training still works and push is automatically disabled.
+Credentials and publish permissions are handled by deployment runtime settings.
 
 ## Runtime controls
 
@@ -45,7 +38,7 @@ If no token is available, public dataset training still works and push is automa
 - `Gate Min pass@1`, `Gate Min pass@k`, `Gate Min Rows`: runtime gate thresholds.
 - `Live Tactical Telemetry`: real-time stage progression, runtime posture, loss sparkline, and gate/push state.
 - `Runtime Terminal Log (Live)`: line-by-line subprocess stream with heartbeat events during silent phases.
-- `Preflight Only (No Training)`: validates pipeline with `--dry-run`.
+- `Validation Mode (No Training)`: validates pipeline with `--dry-run`.
 - `Force Dataset Redownload`: bypasses cached parquet files.
 - `Abort Active Run`: cancels active subprocess tree.
 

app.py

@@ -293,7 +293,7 @@ TACTICAL_HEADER_HTML = """
   <div class="ops-header-title">Maths Conjecture Solutions // Training Operations Console</div>
   <div class="ops-header-tags">
     <span class="ops-tag">Tactical Monochrome</span>
-    <span class="ops-tag">Autonomous Auth</span>
+    <span class="ops-tag">Controlled Ops</span>
     <span class="ops-tag">Staged Curriculum</span>
     <span class="ops-tag">Live Telemetry</span>
   </div>
@@ -311,7 +311,7 @@ An autonomous training operations console for DeepSeek-Math that runs multi-stag
 3. Execute multi-stage DeepSeek-Math curriculum fine-tuning via `scripts/train_sota.py`.
 4. Run post-training evaluation with pass@k-style sampling and family-level metrics.
 5. Enforce autonomous quality gates before adapter promotion/push.
-6. Auto-resolve Hugging Face credentials, stream live telemetry, and emit structured run summaries.
+6. Stream live terminal telemetry, tactical visualization, and structured run summaries.
 """
 
 
@@ -931,7 +931,7 @@ def run_pipeline_core(
         return
 
     try:
-        token, token_source, token_path = resolve_hf_token()
+        token, _, _ = resolve_hf_token()
 
         dataset_repo_id = validate_repo_id(dataset_repo_id, "Dataset repo")
         model_repo_id = validate_repo_id(model_repo_id, "Model repo")
@@ -990,11 +990,6 @@ def run_pipeline_core(
                 "force_redownload": bool(force_redownload),
                 "preflight_only": bool(preflight_only),
                 "runtime": runtime,
-                "auth": {
-                    "token_source": token_source,
-                    "credentials_path": token_path,
-                    "token_present": bool(token),
-                },
             }
         )
 
@@ -1005,12 +1000,9 @@ def run_pipeline_core(
             f"cuda_available={runtime['cuda_available']} devices={runtime['cuda_device_count']}",
         )
         if token:
-            append_log(log_lines, f"Auth resolved from {token_source}.")
+            append_log(log_lines, "Environment posture validated.")
         else:
-            append_log(
-                log_lines,
-                "No HF token found. Continuing without auth; public dataset access works but hub push will be disabled.",
-            )
+            append_log(log_lines, "Restricted mode active. Hub publish disabled for this run.")
         yield "\n".join(log_lines), "Validating environment", summary_text(summary)
 
         if not preflight_only and not torch.cuda.is_available():
@@ -1084,7 +1076,7 @@ def run_pipeline_core(
         ]
         if preflight_only:
             train_cmd.append("--dry-run")
-            append_log(log_lines, "Preflight-only mode enabled: validating stages without training.")
+            append_log(log_lines, "Validation mode enabled: running dry validation without full training.")
 
         train_gen = stream_subprocess(
             cmd=train_cmd,
@@ -1120,7 +1112,7 @@ def run_pipeline_core(
         if preflight_only:
             summary["result"] = "preflight_passed"
             summary["finished_at_utc"] = now_ts()
-            append_log(log_lines, "Preflight checks completed successfully.")
+            append_log(log_lines, "Validation mode completed successfully.")
             yield "\n".join(log_lines), "Preflight complete", summary_text(summary)
             return
 
@@ -1346,7 +1338,7 @@ with gr.Blocks(title="Math Conjecture Trainer Space") as demo:
     with gr.Row():
         push_to_hub = gr.Checkbox(label="Push Adapter to Hub", value=True)
         force_redownload = gr.Checkbox(label="Force Dataset Redownload", value=False)
-        preflight_only = gr.Checkbox(label="Preflight Only (No Training)", value=False)
+        preflight_only = gr.Checkbox(label="Validation Mode (No Training)", value=False)
 
     with gr.Row():
         run_button = gr.Button("Execute Training Run", variant="primary")