Update app.py

app.py

@@ -1,10 +1,10 @@
 # app.py
-import os, shutil, subprocess, zipfile, traceback
+import os, shutil, subprocess, zipfile, traceback, time, io
 from pathlib import Path
 from datetime import datetime
 import gradio as gr
 
-ROOT = Path(__file__).resolve().parent
+ROOT = Path(_file_).resolve().parent
 DATA = ROOT / "dataset.jsonl"
 LOG  = ROOT / "train.log"
 RUNS = ROOT / "runs"
@@ -55,8 +55,12 @@ def upload_dataset(file):
         return f"✅ Uploaded → {DATA.name}", ls_workspace()
     return "⚠ Unexpected item; please upload a .jsonl file.", ls_workspace()
 
-# -------- training --------
-def start_training(run_name):
+# -------- training (LIVE LOGS) --------
+def start_training_live(run_name):
+    """
+    Streams training logs to the UI while the subprocess runs.
+    Yields tuples for outputs: [status, download_file, workspace, logs, model_dropdown]
+    """
     run_id = (run_name or "").strip() or datetime.now().strftime("run_%Y%m%d_%H%M%S")
     out_dir  = RUNS / run_id
     zip_path = RUNS / f"{run_id}.zip"
@@ -67,7 +71,9 @@ def start_training(run_name):
     if zip_path.exists():
         zip_path.unlink()
 
+    # init log
     LOG.write_text(f"🔥 Training started…\nRun: {run_id}\n", encoding="utf-8")
+    append_log(f"Workspace:\n{ls_workspace()}")
 
     cmd = [
         "python", str(ROOT / "train.py"),
@@ -81,21 +87,62 @@ def start_training(run_name):
         "--learning_rate", "5e-5",
     ]
     append_log("▶ " + " ".join(cmd))
-    with open(LOG, "a", encoding="utf-8") as lf:
-        code = subprocess.Popen(cmd, stdout=lf, stderr=subprocess.STDOUT).wait()
+
+    # start subprocess with live stdout
+    proc = subprocess.Popen(
+        cmd,
+        stdout=subprocess.PIPE,
+        stderr=subprocess.STDOUT,
+        bufsize=1,
+        universal_newlines=True,
+        encoding="utf-8",
+        errors="replace",
+    )
+
+    live_log = io.StringIO()
+    status_msg = f"🚀 Training run '{run_id}' in progress…"
+    # stream loop
+    while True:
+        line = proc.stdout.readline()
+        if line == "" and proc.poll() is not None:
+            break
+        if line:
+            append_log(line.rstrip("\n"))
+            live_log.write(line)
+            # Trim to last ~20k chars for UI
+            text = live_log.getvalue()[-20000:]
+            # yield with download hidden (until zip exists)
+            yield (
+                status_msg,
+                gr.update(value=None, visible=False),
+                ls_workspace(),
+                text,
+                dropdown_update_safe(list_models(), prefer=None),
+            )
+            # if zip appears during training (e.g., early save), surface it
+            if zip_path.exists():
+                yield (
+                    "📦 Model zip created during run.",
+                    gr.update(value=str(zip_path), visible=True),
+                    ls_workspace(),
+                    text,
+                    dropdown_update_safe(list_models(), prefer=None),
+                )
+
+    code = proc.wait()
 
     models = list_models()
     model_update = dropdown_update_safe(models, prefer=str(out_dir) if out_dir.exists() else None)
+    final_logs = read_logs()
 
     if code == 0 and zip_path.exists():
         info = f"✅ Training complete. Saved: {out_dir.name} | Zip: {zip_path.name}"
-        dl_update = gr.update(value=str(zip_path), visible=True)
+        append_log(info)
+        yield (info, gr.update(value=str(zip_path), visible=True), ls_workspace(), final_logs, model_update)
     else:
         info = f"❌ Training failed (exit {code}). Check logs below."
-        dl_update = gr.update(value=None, visible=False)
-
-    append_log(info)
-    return info, dl_update, ls_workspace(), read_logs(), model_update
+        append_log(info)
+        yield (info, gr.update(value=None, visible=False), ls_workspace(), final_logs, model_update)
 
 def refresh_download():
     zips = sorted(RUNS.glob("*.zip"), key=lambda p: p.stat().st_mtime, reverse=True)
@@ -165,6 +212,7 @@ def generate(model_path, prompt):
     if not model_path:
         return "❌ Select a model from the dropdown first."
     if not isinstance(model_path, str):
+        # fix minor bug: _name_ not name
         return f"❌ Invalid model path type: {type(model_path)._name_}"
     if not Path(model_path).exists():
         return f"❌ Model folder not found: {model_path}"
@@ -193,7 +241,7 @@ def generate(model_path, prompt):
 
 # -------- UI --------
 with gr.Blocks(title="Python AI — Train & Test") as app:
-    gr.Markdown("## 🧠 Python AI — Train & Test\n• Unique runs • Safe download • Cached generation\n")
+    gr.Markdown("## 🧠 Python AI — Train & Test\n• Unique runs • Safe download • Cached generation • Live logs\n")
 
     # Test first (so Train can update its dropdown)
     with gr.Tab("Test"):
@@ -218,19 +266,22 @@ with gr.Blocks(title="Python AI — Train & Test") as app:
             ws = gr.Textbox(label="Workspace", lines=16, value=ls_workspace())
         run_name = gr.Textbox(label="Run name (optional)", placeholder="e.g., python_small_v1")
         up_status = gr.Textbox(label="Upload Status", interactive=False)
-        start = gr.Button("🚀 Start Training", variant="primary")
-        logs = gr.Textbox(label="📜 Training Logs", lines=18, value=read_logs())
+        start = gr.Button("🚀 Start Training (Live Logs)", variant="primary")
+        logs = gr.Textbox(label="📜 Training Logs (live)", lines=18, value=read_logs())
         status = gr.Textbox(label="Status", interactive=False)
         download_file = gr.File(label="📦 Latest trained zip", visible=False)
         refresh_dl_btn = gr.Button("Refresh Download")
 
     # wiring
     ds.change(upload_dataset, inputs=ds, outputs=[up_status, ws])
+
+    # STREAMED training: function yields updates
     start.click(
-        start_training,
+        start_training_live,
         inputs=[run_name],
         outputs=[status, download_file, ws, logs, model_list]
     )
+
     refresh_dl_btn.click(
         refresh_download,
         outputs=[download_file, ws, model_list]