scripts/deprecated/upload_v6_5_7ds.py

"""upload_v6_5_7ds.py — V6.5-7ds upload to HuggingFace.

User requirement (latest): "upload dos arquivos testados e os aprimorados
corrigidos (remover os antigos)"

This script:
  1. Pre-upload safety: scrub HF_TOKEN from all scripts/*.py via regex.
  2. Defines OLD_FILES_TO_REMOVE — legacy v3/v4/v5/v6_1/v6_2/v6_3/v6_progressive
     reports and scripts, plus old v6_5_* files (superseded by v6_5_7ds_*).
  3. Uploads the tested+improved files (current BiGRU_T_version tree).
  4. Deletes OLD_FILES_TO_REMOVE from the HF repo via HfApi.delete_file.
  5. Verifies critical V6.5-7ds files are present in repo.
  6. Saves upload report to v6_5_7ds_upload_report.json.

After successful upload, the HF_TOKEN env var MUST be deleted by the caller.
"""
from __future__ import annotations
import json, os, sys, time, re
from pathlib import Path

PROJECT_ROOT = Path("/home/z/my-project")
BIGRU_ROOT = PROJECT_ROOT / "BiGRU_T_version"

# ──────────────────────────────────────────────────────────────────────────────
# Files to REMOVE from the HF repo (legacy / superseded by V6.5-7ds)
# ──────────────────────────────────────────────────────────────────────────────
# User requirement: "upload dos arquivos testados e os aprimorados corrigidos
# (remover os antigos)"
#
# Removal strategy:
# - All legacy version reports (v3, v4, v5, v6, v6_1, v6_2, v6_3, v6_progressive)
# - All legacy upload reports (v3, v5, v6_1, v6_2, v6_4, v6_5)
# - Old V6.5 files superseded by V6.5-7ds (v6_5_*.json and v6_5_final_*.json)
# - Old training scripts (train.py, train_v2.py, train_fast.py, smoke_test.py,
#   parse_v6_log.py, train_v6_1.py, train_v6_2.py, train_v6_3.py,
#   train_v6_progressive.py, train_v6_5.py, train_v6_5_final.py)
# - Old upload scripts (upload_to_hf.py, upload_v6_1_resilient.py,
#   upload_v6_2_resilient.py, upload_v6_3_resilient.py, upload_v6_4_resilient.py,
#   upload_v6_5_resilient.py)
# - Legacy root files: bug_hunt_v3_report.json, v32_test_report.json,
#   v32_upload_report.json, training_report.json, v6_progressive_train.log,
#   v6_progressive_report.json, v6_report.json, v6_upload_report.json,
#   config.json (legacy model_final/config.json stays)
OLD_FILES_TO_REMOVE = [
    # ── Legacy root reports (v3/v4/v5/v6/v32) ─────────────────────────────────
    "bug_hunt_v3_report.json",
    "v3_report.json",
    "v4_report.json",
    "v5_report.json",
    "v5_upload_report.json",
    "v32_test_report.json",
    "v32_upload_report.json",
    "training_report.json",
    "v6_report.json",
    "v6_upload_report.json",
    "v6_progressive_report.json",
    "v6_progressive_train.log",
    # ── Legacy v6.1/v6.2/v6.3 reports ─────────────────────────────────────────
    "v6_1_report.json",
    "v6_1_upload_report.json",
    "v6_2_report.json",
    "v6_2_upload_report.json",
    "v6_3_report.json",
    "v6_3_training_metrics.json",
    # ── Old V6.5 reports (superseded by V6.5-7ds) ─────────────────────────────
    # These are removed because V6.5-7ds is the new canonical version.
    "v6_5_report.json",
    "v6_5_training_metrics.json",
    "v6_5_module_analysis.json",
    "v6_5_script_activity.json",
    "v6_5_ewc_w8a8_benchmark.json",
    "v6_5_reasoning_eval.json",
    "v6_5_upload_report.json",
    # ── Old V6.5-final reports (also superseded by V6.5-7ds) ──────────────────
    "v6_5_final_report.json",
    "v6_5_final_training_metrics.json",
    "v6_5_final_module_analysis.json",
    "v6_5_final_script_activity.json",
    "v6_5_final_ewc_w8a8_benchmark.json",
    "v6_5_final_reasoning_eval.json",
    "v6_5_final_w8a8_compression.json",
    # ── Legacy training scripts ───────────────────────────────────────────────
    "scripts/parse_v6_log.py",
    "scripts/smoke_test.py",
    "scripts/train.py",
    "scripts/train_fast.py",
    "scripts/train_v2.py",
    "scripts/train_v6_1.py",
    "scripts/train_v6_2.py",
    "scripts/train_v6_3.py",
    "scripts/train_v6_progressive.py",
    "scripts/train_v6_5.py",            # superseded by train_v6_5_7ds.py
    "scripts/train_v6_5_final.py",      # superseded by train_v6_5_7ds.py
    # ── Legacy upload scripts (only upload_v6_5_7ds.py stays) ─────────────────
    "scripts/upload_to_hf.py",
    "scripts/upload_v6_1_resilient.py",
    "scripts/upload_v6_2_resilient.py",
    "scripts/upload_v6_3_resilient.py",
    "scripts/upload_v6_4_resilient.py",
    "scripts/upload_v6_5_resilient.py",  # superseded by upload_v6_5_7ds.py
    # ── Legacy tokenizer/ root config (kept model_final/ tree intact) ─────────
    "config.json",                       # legacy root config; not used by V6.5
    "tokenizer/tokenizer.json",          # legacy root tokenizer; model_final/ has its own
    # ── Legacy data_augmentation.py (was moved to training/ in V6.4) ──────────
    "src/bigru_t/data/data_augmentation.py",
    # ── Legacy xeon_runtime.py at scripts/ (canonical is utils/xeon_runtime.py)
    "scripts/xeon_runtime.py",
]

# Critical files that MUST be present after upload
CRITICAL_FILES_V65_7DS = [
    # Core model
    "src/bigru_t/model/kohonen_learning_system.py",
    "src/bigru_t/model/__init__.py",
    "src/bigru_t/__init__.py",
    "src/bigru_t/model/hyp_t.py",
    "src/bigru_t/model/vqvae2_hierarchical.py",
    "src/bigru_t/model/vqvae2_hierarchical_flexnet.py",
    "src/bigru_t/model/token_compress.py",
    "src/bigru_t/model/embedding_reconfig.py",
    "src/bigru_t/model/attention_multimodal.py",
    # Training
    "src/bigru_t/training/mtp.py",
    "src/bigru_t/training/ewc.py",
    # Quantization
    "src/bigru_t/quantization/smoothquant_compressor.py",
    "src/bigru_t/quantization/w8a8_smoothquant.py",
    "src/bigru_t/quantization/quantized_linear.py",
    # Reasoning
    "src/bigru_t/reasoning/thinking.py",
    "src/bigru_t/reasoning/reasoning_engine.py",
    "src/bigru_t/reasoning/circular_orchestration.py",
    "src/bigru_t/reasoning/tool_agent.py",
    "src/bigru_t/reasoning/distributed_reasoning_system.py",
    "src/bigru_t/reasoning/cyclic_reasoning.py",
    "src/bigru_t/reasoning/consensus_sampling.py",
    # Data + utils
    "src/bigru_t/data/streaming_datasets.py",
    "src/bigru_t/utils/xeon_runtime.py",
    # V6.5-7ds reports (NEW canonical)
    "v6_5_7ds_report.json",
    "v6_5_7ds_training_metrics.json",
    "v6_5_7ds_module_analysis.json",
    "v6_5_7ds_script_activity.json",
    "v6_5_7ds_ewc_w8a8_benchmark.json",
    "v6_5_7ds_reasoning_eval.json",
    "v6_5_7ds_w8a8_compression.json",
    "v6_5_7ds_upload_report.json",
    # V6.4 reports (kept as the immediate predecessor baseline)
    "v6_4_report.json",
    "v6_4_training_metrics.json",
    "v6_4_upload_report.json",
    # V6.5-7ds scripts (the new canonicals)
    "scripts/train_v6_5_7ds.py",
    "scripts/upload_v6_5_7ds.py",
    "scripts/train_v6_4.py",  # kept as predecessor reference
    # Misc
    "requirements.txt",
    "README.md",
    "docs/analysis.md",
]


def main() -> int:
    print("=" * 72)
    print("V6.5-7DS — UPLOAD TO HUGGINGFACE (with removal of old files)")
    print("=" * 72)
    hf_token = os.environ.get("HF_TOKEN")
    if not hf_token:
        print("ERROR: HF_TOKEN not set in env")
        return 1
    print(f"  HF_TOKEN loaded from env (length={len(hf_token)})")

    # ── Step 1: Pre-upload safety — scrub any HF token from scripts ──────────
    token_pattern = re.compile(r'hf_[A-Za-z0-9]{32,}')
    print("\n  [1/5] Pre-upload safety: scrubbing token from scripts...")
    scrubbed_count = 0
    for script_path in BIGRU_ROOT.glob("scripts/*.py"):
        try:
            content = script_path.read_text()
            if token_pattern.search(content):
                scrubbed = token_pattern.sub("hf_<REDACTED_TOKEN>", content)
                script_path.write_text(scrubbed)
                scrubbed_count += 1
                print(f"    SCRUBBED: {script_path.name}")
        except Exception as e:
            print(f"    SKIP {script_path.name}: {e}")
    if scrubbed_count == 0:
        print("    (no token leakage found in scripts)")

    # ── Step 2: Import HF API + check repo access ────────────────────────────
    print("\n  [2/5] Connecting to HuggingFace repo...")
    try:
        from huggingface_hub import HfApi, upload_folder
    except ImportError:
        print("ERROR: huggingface_hub not installed")
        return 1

    api = HfApi(token=hf_token)
    repo_id = "PowerMachine/BiGRU_T_version"
    try:
        info = api.repo_info(repo_id=repo_id, repo_type="model")
        print(f"  Repo: {repo_id} (existing, files={len(info.siblings)})")
    except Exception as e:
        print(f"  ERROR: cannot access repo: {e}")
        return 1

    # Snapshot of files in repo BEFORE upload
    files_before = set(s.rfilename for s in info.siblings)
    print(f"  Files in repo BEFORE upload: {len(files_before)}")

    # ── Step 3: Upload current BiGRU_T_version tree as single commit ─────────
    print(f"\n  [3/5] Uploading {BIGRU_ROOT} as single commit...")
    t0 = time.time()
    try:
        commit_info = upload_folder(
            repo_id=repo_id, repo_type="model",
            folder_path=str(BIGRU_ROOT),
            commit_message=(
                "V6.5-7ds: 7 datasets streaming (1400 samples, 112 steps), "
                "864 neurons, SmoothQuant W8A8 (err=0.022), MTP K=6, "
                "tool_coordinator workers reactivated, 10/10 verification PASS, "
                "reasoning GOOD, removed legacy v3/v4/v5/v6_1/v6_2/v6_3/v6_progressive "
                "files + old v6_5_* reports/scripts"
            ),
            token=hf_token,
        )
        t1 = time.time()
        print(f"  Upload completed in {t1 - t0:.2f}s")
        print(f"  Commit: {commit_info.oid}")
        print(f"  URL: https://huggingface.co/PowerMachine/BiGRU_T_version/commit/{commit_info.oid}")
    except Exception as e:
        print(f"  ERROR: upload failed: {e}")
        return 1

    # ── Step 4: Remove OLD files from repo (single delete commit) ────────────
    print(f"\n  [4/5] Removing {len(OLD_FILES_TO_REMOVE)} old files from repo...")
    # Refresh repo info to know what files currently exist
    try:
        info_after = api.repo_info(repo_id=repo_id, repo_type="model")
        files_after_upload = set(s.rfilename for s in info_after.siblings)
    except Exception as e:
        print(f"  WARNING: cannot refresh repo info: {e}")
        files_after_upload = files_before

    # Determine which old files actually exist in repo (to delete)
    old_files_in_repo = [f for f in OLD_FILES_TO_REMOVE if f in files_after_upload]
    old_files_not_in_repo = [f for f in OLD_FILES_TO_REMOVE if f not in files_after_upload]
    print(f"    Old files present in repo (will delete): {len(old_files_in_repo)}")
    print(f"    Old files NOT in repo (skip): {len(old_files_not_in_repo)}")
    if old_files_not_in_repo:
        print(f"      examples: {old_files_not_in_repo[:5]}")

    # Use HfApi.create_commit with CommitOperationDelete for batch removal
    # (more efficient than calling delete_file one-by-one)
    delete_errors = []
    if old_files_in_repo:
        try:
            from huggingface_hub import CommitOperation
            operations = [
                CommitOperationDelete(path_in_repo=f) for f in old_files_in_repo
            ]
            print(f"    Creating batch delete commit ({len(operations)} files)...")
            t_del_start = time.time()
            commit_del = api.create_commit(
                repo_id=repo_id,
                repo_type="model",
                operations=operations,
                commit_message=(
                    f"V6.5-7ds cleanup: remove {len(operations)} legacy/superseded files "
                    f"(v3/v4/v5/v6_1/v6_2/v6_3/v6_progressive + old v6_5_*/v6_5_final_* "
                    f"reports/scripts) — replaced by V6.5-7ds canonicals"
                ),
            )
            t_del_end = time.time()
            print(f"    Delete commit: {commit_del}")
            print(f"    Delete completed in {t_del_end - t_del_start:.2f}s")
            print(f"    URL: https://huggingface.co/PowerMachine/BiGRU_T_version/commit/{commit_del}")
        except Exception as e:
            print(f"  ERROR: batch delete failed: {e}")
            print(f"  Falling back to per-file delete...")
            # Fallback: delete one-by-one
            for f in old_files_in_repo:
                try:
                    api.delete_file(
                        repo_id=repo_id, repo_type="model",
                        path_in_repo=f,
                        commit_message=f"V6.5-7ds cleanup: remove legacy {f}",
                        token=hf_token,
                    )
                    print(f"    DELETED: {f}")
                except Exception as e2:
                    delete_errors.append((f, str(e2)[:100]))
                    print(f"    FAILED: {f} — {str(e2)[:100]}")

    # ── Step 5: Verify critical files in repo ────────────────────────────────
    print(f"\n  [5/5] Verifying critical V6.5-7ds files in repo...")
    try:
        # Refresh repo info after deletes
        info_final = api.repo_info(repo_id=repo_id, repo_type="model")
        files_in_repo = set(s.rfilename for s in info_final.siblings)
        all_present = True
        for cf in CRITICAL_FILES_V65_7DS:
            present = cf in files_in_repo
            mark = "OK" if present else "MISS"
            print(f"    [{mark}] {cf}")
            if not present:
                all_present = False
        print(f"\n  Files in repo AFTER cleanup: {len(files_in_repo)}")
        if not all_present:
            print("  WARNING: some critical files missing!")
    except Exception as e:
        print(f"  WARNING: cannot verify files: {e}")
        files_in_repo = set()

    # ── Save upload report ───────────────────────────────────────────────────
    report = {
        "version": "V6.5-7ds",
        "upload_timestamp": time.strftime("%Y-%m-%dT%H:%M:%S"),
        "repo_id": repo_id,
        "upload_commit_oid": commit_info.oid,
        "upload_commit_url": f"https://huggingface.co/PowerMachine/BiGRU_T_version/commit/{commit_info.oid}",
        "upload_duration_s": float(t1 - t0),
        "delete_commit_oid": commit_del if old_files_in_repo else None,
        "delete_count": len(old_files_in_repo) if old_files_in_repo else 0,
        "delete_errors": delete_errors,
        "n_files_before": len(files_before),
        "n_files_after_upload": len(files_after_upload),
        "n_files_after_cleanup": len(files_in_repo) if files_in_repo else None,
        "old_files_removed": old_files_in_repo,
        "old_files_not_in_repo_skipped": old_files_not_in_repo,
        "critical_files": CRITICAL_FILES_V65_7DS,
        "all_critical_files_present": all_present if 'all_present' in dir() else None,
        "token_scrubbed_count": scrubbed_count,
    }
    report_path = BIGRU_ROOT / "v6_5_7ds_upload_report.json"
    report_path.write_text(json.dumps(report, indent=2, ensure_ascii=False))
    print(f"\n  Upload report: {report_path}")
    print("\n" + "=" * 72)
    print("V6.5-7DS — UPLOAD COMPLETED (with old file removal)")
    print("=" * 72)
    return 0


if __name__ == "__main__":
    sys.exit(main())