f

.claude/settings.local.json

@@ -32,7 +32,9 @@
       "Bash(where magick *)",
       "Bash(where inkscape *)",
       "Bash(python -c \"import cairosvg; cairosvg.svg2png\\(url='pipeline_diagram.svg', write_to='pipeline_diagram.png', output_width=2800\\)\")",
-      "Bash(python scripts/_patch_notebook.py)"
+      "Bash(python scripts/_patch_notebook.py)",
+      "Bash(python scripts/_patch_notebook2.py)",
+      "Bash(python scripts/_verify_nb.py)"
     ]
   }
 }

.claude/worktrees/keen-galileo-ce978c

@@ -0,0 +1 @@
+Subproject commit 28b13fcded5614d6ab4088a8d8f6a20b87060306

.claude/worktrees/nice-kapitsa-0be5ac

@@ -0,0 +1 @@
+Subproject commit 28b13fcded5614d6ab4088a8d8f6a20b87060306

count_img_on_HF.py

@@ -0,0 +1,62 @@
+from huggingface_hub import HfApi
+from huggingface_hub.hf_api import RepoFile
+
+# ── Sửa 2 dòng này theo nhu cầu ──────────────────────────────────────────────
+REPO_ID = "hieu3636/cxr-vlm-data"   # tên repo HuggingFace
+FOLDER  = "IU-Xray_2"
+# folder cần đếm, ví dụ "files/p10" hoặc None để scan toàn bộ
+# ─────────────────────────────────────────────────────────────────────────────
+
+IMAGE_EXTENSIONS = {".jpg", ".jpeg", ".png", ".bmp", ".gif", ".tiff", ".tif", ".webp", ".dcm"}
+
+
+def fmt_size(n_bytes: float) -> str:
+    for unit in ("B", "KB", "MB", "GB", "TB"):
+        if n_bytes < 1024:
+            return f"{n_bytes:.2f} {unit}"
+        n_bytes /= 1024
+    return f"{n_bytes:.2f} PB"
+
+
+def main() -> None:
+    api = HfApi()
+
+    print(f"Repo    : {REPO_ID}")
+    print(f"Folder  : {FOLDER or '(toàn bộ repo)'}")
+    print("Đang lấy danh sách file...\n")
+
+    total_images = 0
+    total_bytes  = 0
+    ext_counts: dict[str, int] = {}
+
+    for item in api.list_repo_tree(
+        repo_id=REPO_ID,
+        repo_type="dataset",
+        path_in_repo=FOLDER or "",
+        recursive=True,
+        expand=True,        # lấy kèm metadata (size, ...)
+    ):
+        if not isinstance(item, RepoFile):
+            continue
+        name = item.rfilename
+        ext = ("." + name.rsplit(".", 1)[-1].lower()) if "." in name else ""
+        if ext not in IMAGE_EXTENSIONS:
+            continue
+        size = item.size or 0
+        total_images += 1
+        total_bytes  += size
+        ext_counts[ext] = ext_counts.get(ext, 0) + 1
+
+    if total_images % 1000 == 0:
+            print(f"  ... đã đếm {total_images:,} ảnh", flush=True)
+
+    print(f"\nTổng số ảnh    : {total_images:,}")
+    print(f"Tổng dung lượng: {fmt_size(total_bytes)}")
+    if ext_counts:
+        print("\nTheo đuôi file:")
+        for ext, cnt in sorted(ext_counts.items(), key=lambda x: -x[1]):
+            print(f"  {ext:10s} {cnt:,}")
+
+
+if __name__ == "__main__":
+    main()

evaluation/metrics.py

@@ -251,7 +251,9 @@ def evaluate_all(
     """
     results = {}
 
-    if task in ("findings", "impression"):
+    # "report" is the merged-mode task (full Findings + Impression in one
+    # target). Same NLG/clinical metrics apply as for findings/impression.
+    if task in ("findings", "impression", "report"):
         results.update(compute_bleu(hypotheses, references))
         results.update(compute_rouge(hypotheses, references))
         results.update(compute_bertscore(hypotheses, references, device=device))

scripts/cxrvlm_colab_train.ipynb

@@ -1138,6 +1138,44 @@
     "from datetime import datetime as _dt, timezone as _tz\n",
     "from pathlib import Path as _Path\n",
     "\n",
+    "def _pull_timing_from_hf(run_id, ckpt_root, repo_id, token):\n",
+    "    # Pull timing.json from HF Hub for this run if not present locally.\n",
+    "    local = ckpt_root / run_id / \"timing.json\"\n",
+    "    if local.exists() or not repo_id or not token:\n",
+    "        return\n",
+    "    try:\n",
+    "        from huggingface_hub import hf_hub_download\n",
+    "        hf_hub_download(\n",
+    "            repo_id   = repo_id,\n",
+    "            repo_type = \"model\",\n",
+    "            filename  = f\"{run_id}/timing.json\",\n",
+    "            token     = token,\n",
+    "            local_dir = str(ckpt_root),\n",
+    "        )\n",
+    "        print(f\"[TIMING] pulled previous timing.json from HF → {local}\")\n",
+    "    except Exception as e:\n",
+    "        # First run for this run_id → no remote file yet. That's fine.\n",
+    "        pass\n",
+    "\n",
+    "def _push_timing_to_hf(run_id, ckpt_root, repo_id, token):\n",
+    "    # Upload local timing.json to HF Hub under {run_id}/timing.json.\n",
+    "    local = ckpt_root / run_id / \"timing.json\"\n",
+    "    if not local.exists() or not repo_id or not token:\n",
+    "        return\n",
+    "    try:\n",
+    "        from huggingface_hub import HfApi\n",
+    "        HfApi(token=token).upload_file(\n",
+    "            path_or_fileobj = str(local),\n",
+    "            path_in_repo   = f\"{run_id}/timing.json\",\n",
+    "            repo_id        = repo_id,\n",
+    "            repo_type      = \"model\",\n",
+    "            commit_message = f\"timing.json @ {run_id}\",\n",
+    "        )\n",
+    "        print(f\"[TIMING] uploaded timing.json to HF → {repo_id}/{run_id}/timing.json\")\n",
+    "    except Exception as e:\n",
+    "        print(f\"[TIMING] upload failed (non-fatal): {e}\")\n",
+    "\n",
+    "\n",
     "_resume_args = \"\"\n",
     "_is_resume = False\n",
     "if \"RESUME_FROM\" in dir() and RESUME_FROM and RESUME_STAGE == 1:\n",
@@ -1147,6 +1185,12 @@
     "else:\n",
     "    print(\"▶ STAGE 1 fresh run\")\n",
     "\n",
+    "# ─── Pre-pull timing.json from HF if resuming (best-effort) ────────────────\n",
+    "_hf_repo  = getattr(train_cfg.hf_hub, \"repo_id\", None) if train_cfg.hf_hub.enabled else None\n",
+    "_hf_token = os.environ.get(\"HF_TOKEN\")\n",
+    "if _is_resume and \"RESUME_RUN_ID\" in dir() and RESUME_RUN_ID:\n",
+    "    _pull_timing_from_hf(RESUME_RUN_ID, CKPT_ROOT, _hf_repo, _hf_token)\n",
+    "\n",
     "# ─── Start-of-stage timer ──────────────────────────────────────────────────\n",
     "_t0_stage1 = _time.time()\n",
     "_iso_start_stage1 = _dt.now(_tz.utc).isoformat(timespec=\"seconds\")\n",
@@ -1187,6 +1231,9 @@
     "    })\n",
     "    _timing_path.write_text(_json.dumps(_t, indent=2))\n",
     "\n",
+    "    # ─── Push to HF Hub so the timer survives a fresh VM ─────────────────\n",
+    "    _push_timing_to_hf(_run_id_now, CKPT_ROOT, _hf_repo, _hf_token)\n",
+    "\n",
     "    def _fmt(sec):\n",
     "        h, r = divmod(int(sec), 3600); m, s = divmod(r, 60); return f\"{h:d}h {m:02d}m {s:02d}s\"\n",
     "    print()\n",
@@ -1233,6 +1280,44 @@
     "from datetime import datetime as _dt, timezone as _tz\n",
     "from pathlib import Path as _Path\n",
     "\n",
+    "def _pull_timing_from_hf(run_id, ckpt_root, repo_id, token):\n",
+    "    # Pull timing.json from HF Hub for this run if not present locally.\n",
+    "    local = ckpt_root / run_id / \"timing.json\"\n",
+    "    if local.exists() or not repo_id or not token:\n",
+    "        return\n",
+    "    try:\n",
+    "        from huggingface_hub import hf_hub_download\n",
+    "        hf_hub_download(\n",
+    "            repo_id   = repo_id,\n",
+    "            repo_type = \"model\",\n",
+    "            filename  = f\"{run_id}/timing.json\",\n",
+    "            token     = token,\n",
+    "            local_dir = str(ckpt_root),\n",
+    "        )\n",
+    "        print(f\"[TIMING] pulled previous timing.json from HF → {local}\")\n",
+    "    except Exception as e:\n",
+    "        # First run for this run_id → no remote file yet. That's fine.\n",
+    "        pass\n",
+    "\n",
+    "def _push_timing_to_hf(run_id, ckpt_root, repo_id, token):\n",
+    "    # Upload local timing.json to HF Hub under {run_id}/timing.json.\n",
+    "    local = ckpt_root / run_id / \"timing.json\"\n",
+    "    if not local.exists() or not repo_id or not token:\n",
+    "        return\n",
+    "    try:\n",
+    "        from huggingface_hub import HfApi\n",
+    "        HfApi(token=token).upload_file(\n",
+    "            path_or_fileobj = str(local),\n",
+    "            path_in_repo   = f\"{run_id}/timing.json\",\n",
+    "            repo_id        = repo_id,\n",
+    "            repo_type      = \"model\",\n",
+    "            commit_message = f\"timing.json @ {run_id}\",\n",
+    "        )\n",
+    "        print(f\"[TIMING] uploaded timing.json to HF → {repo_id}/{run_id}/timing.json\")\n",
+    "    except Exception as e:\n",
+    "        print(f\"[TIMING] upload failed (non-fatal): {e}\")\n",
+    "\n",
+    "\n",
     "_resume_args = \"\"\n",
     "_is_resume = False\n",
     "if \"RESUME_FROM\" in dir() and RESUME_FROM and RESUME_STAGE == 2:\n",
@@ -1243,7 +1328,24 @@
     "    _resume_args = f'--run_id \"{RESUME_RUN_ID}\"'\n",
     "    print(\"▶ STAGE 2 fresh start, pinned to run_id\", RESUME_RUN_ID)\n",
     "else:\n",
-    "    print(\"▶ STAGE 2 fresh (same session as stage1)\")\n",
+    "    # ─── FIX: pin stage 2 to the run_id stage 1 just wrote ────────────────\n",
+    "    # Without this, train.py treats stage 2 as a brand-new launch and\n",
+    "    # allocates a NEW run_N folder, splitting stage1/stage2 across two runs.\n",
+    "    _state_file = CKPT_ROOT / \"run_id.txt\"\n",
+    "    if _state_file.exists():\n",
+    "        _pinned = _state_file.read_text().strip()\n",
+    "        _resume_args = f'--run_id \"{_pinned}\"'\n",
+    "        print(f\"▶ STAGE 2 fresh, auto-pinned to run_id from state file: {_pinned}\")\n",
+    "    else:\n",
+    "        print(\"▶ STAGE 2 fresh (no state file — train.py will allocate a new run_id)\")\n",
+    "\n",
+    "# ─── Pre-pull timing.json from HF (in case of fresh VM) ───────────────────\n",
+    "_hf_repo  = getattr(train_cfg.hf_hub, \"repo_id\", None) if train_cfg.hf_hub.enabled else None\n",
+    "_hf_token = os.environ.get(\"HF_TOKEN\")\n",
+    "# Best guess at run_id BEFORE training (may be missing if stage 1 wasn't run here)\n",
+    "_pre_state = CKPT_ROOT / \"run_id.txt\"\n",
+    "if _pre_state.exists():\n",
+    "    _pull_timing_from_hf(_pre_state.read_text().strip(), CKPT_ROOT, _hf_repo, _hf_token)\n",
     "\n",
     "# ─── Start-of-stage timer ──────────────────────────────────────────────────\n",
     "_t0_stage2 = _time.time()\n",
@@ -1285,6 +1387,9 @@
     "    })\n",
     "    _timing_path.write_text(_json.dumps(_t, indent=2))\n",
     "\n",
+    "    # ─── Push to HF Hub ──────────────────────────────────────────────────\n",
+    "    _push_timing_to_hf(_run_id_now, CKPT_ROOT, _hf_repo, _hf_token)\n",
+    "\n",
     "    def _fmt(sec):\n",
     "        h, r = divmod(int(sec), 3600); m, s = divmod(r, 60); return f\"{h:d}h {m:02d}m {s:02d}s\"\n",
     "    _total = _t[\"stage1_elapsed_sec\"] + _t[\"stage2_elapsed_sec\"]\n",
@@ -1444,11 +1549,58 @@
     "import json as _json\n",
     "from pathlib import Path as _Path\n",
     "\n",
+    "def _pull_timing_from_hf(run_id, ckpt_root, repo_id, token):\n",
+    "    # Pull timing.json from HF Hub for this run if not present locally.\n",
+    "    local = ckpt_root / run_id / \"timing.json\"\n",
+    "    if local.exists() or not repo_id or not token:\n",
+    "        return\n",
+    "    try:\n",
+    "        from huggingface_hub import hf_hub_download\n",
+    "        hf_hub_download(\n",
+    "            repo_id   = repo_id,\n",
+    "            repo_type = \"model\",\n",
+    "            filename  = f\"{run_id}/timing.json\",\n",
+    "            token     = token,\n",
+    "            local_dir = str(ckpt_root),\n",
+    "        )\n",
+    "        print(f\"[TIMING] pulled previous timing.json from HF → {local}\")\n",
+    "    except Exception as e:\n",
+    "        # First run for this run_id → no remote file yet. That's fine.\n",
+    "        pass\n",
+    "\n",
+    "def _push_timing_to_hf(run_id, ckpt_root, repo_id, token):\n",
+    "    # Upload local timing.json to HF Hub under {run_id}/timing.json.\n",
+    "    local = ckpt_root / run_id / \"timing.json\"\n",
+    "    if not local.exists() or not repo_id or not token:\n",
+    "        return\n",
+    "    try:\n",
+    "        from huggingface_hub import HfApi\n",
+    "        HfApi(token=token).upload_file(\n",
+    "            path_or_fileobj = str(local),\n",
+    "            path_in_repo   = f\"{run_id}/timing.json\",\n",
+    "            repo_id        = repo_id,\n",
+    "            repo_type      = \"model\",\n",
+    "            commit_message = f\"timing.json @ {run_id}\",\n",
+    "        )\n",
+    "        print(f\"[TIMING] uploaded timing.json to HF → {repo_id}/{run_id}/timing.json\")\n",
+    "    except Exception as e:\n",
+    "        print(f\"[TIMING] upload failed (non-fatal): {e}\")\n",
+    "\n",
+    "\n",
     "_run_id_file = CKPT_ROOT / \"run_id.txt\"\n",
     "assert _run_id_file.exists(), \"No run_id.txt — train at least one stage first.\"\n",
     "_run_id = _run_id_file.read_text().strip()\n",
+    "\n",
+    "# Pull the latest timing.json from HF in case we're on a fresh VM.\n",
+    "_hf_repo  = getattr(train_cfg.hf_hub, \"repo_id\", None) if train_cfg.hf_hub.enabled else None\n",
+    "_hf_token = os.environ.get(\"HF_TOKEN\")\n",
+    "_pull_timing_from_hf(_run_id, CKPT_ROOT, _hf_repo, _hf_token)\n",
+    "\n",
     "_timing_path = CKPT_ROOT / _run_id / \"timing.json\"\n",
-    "assert _timing_path.exists(), f\"No timing.json under {_timing_path.parent} — was the stage cell run via the wrapped version?\"\n",
+    "assert _timing_path.exists(), (\n",
+    "    f\"No timing.json under {_timing_path.parent} (also not on HF). \"\n",
+    "    f\"Was the stage cell run via the wrapped version?\"\n",
+    ")\n",
     "\n",
     "_t = _json.loads(_timing_path.read_text())\n",
     "\n",