1# /// script
2# dependencies = [
3#   "openai>=1.0",
4#   "azure-identity",
5#   "azure-ai-projects",
6# ]
7# ///
8"""
9cleanup.py — Clean up fine-tuning resources to avoid quota exhaustion.
10 
11Lists and optionally deletes uploaded files and cancels pending jobs.
12Useful after experimentation to reclaim quota (max 100 files per resource,
13deployment slots are limited).
14 
15Usage:
16  python cleanup.py --list                            # List all resources
17  python cleanup.py --list --type files               # List only files
18  python cleanup.py --delete-files --older-than 7     # Delete files older than 7 days
19  python cleanup.py --delete-files --dry-run          # Preview what would be deleted
20  python cleanup.py --cancel-pending                  # Cancel queued jobs
21"""
22 
23import argparse
24import os
25import sys
26 
27try:
28    sys.stdout.reconfigure(encoding="utf-8")
29    sys.stderr.reconfigure(encoding="utf-8")
30except (AttributeError, OSError):
31    pass  # Stream not reconfigurable (older Python or non-tty); default encoding is fine
32from datetime import datetime, timezone
33 
34sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
35from common import HelpOnErrorParser, get_clients
36 
37 
38def list_deployments(client):
39    """List fine-tuned model deployments. Returns deployment info from jobs."""
40    deployments = []
41    for job in _iter_all_jobs(client):
42        if job.fine_tuned_model and job.status == "succeeded":
43            deployments.append({
44                "job_id": job.id,
45                "model": job.fine_tuned_model,
46                "base_model": job.model,
47                "created": datetime.fromtimestamp(job.created_at, tz=timezone.utc),
48                "tokens": job.trained_tokens,
49            })
50    return deployments
51 
52 
53def _iter_all_jobs(client, page_size=100):
54    """Yield every fine-tuning job, paginating through the API.
55 
56    The OpenAI/Azure SDK's `jobs.list(limit=N)` returns at most N jobs with no
57    auto-paging. Users with >100 jobs would otherwise miss older jobs entirely.
58    """
59    after = None
60    while True:
61        kwargs = {"limit": page_size}
62        if after:
63            kwargs["after"] = after
64        page = client.fine_tuning.jobs.list(**kwargs)
65        items = list(page)
66        if not items:
67            break
68        for job in items:
69            yield job
70        if len(items) < page_size:
71            break
72        # Cursor-based paging: use last job's id as `after`
73        after = items[-1].id
74 
75 
76def list_files(client):
77    """List uploaded files."""
78    files = client.files.list()
79    result = []
80    for f in files:
81        result.append({
82            "id": f.id,
83            "filename": f.filename,
84            "purpose": f.purpose,
85            "bytes": f.bytes,
86            "created": datetime.fromtimestamp(f.created_at, tz=timezone.utc),
87            "status": f.status,
88        })
89    return result
90 
91 
92def list_jobs(client):
93    """List fine-tuning jobs."""
94    result = []
95    for job in _iter_all_jobs(client):
96        result.append({
97            "id": job.id,
98            "status": job.status,
99            "model": job.model,
100            "fine_tuned_model": job.fine_tuned_model or "—",
101            "created": datetime.fromtimestamp(job.created_at, tz=timezone.utc),
102        })
103    return result
104 
105 
106def format_age(dt):
107    """Format a datetime as a human-readable age."""
108    delta = datetime.now(timezone.utc) - dt
109    if delta.days > 0:
110        return f"{delta.days}d ago"
111    hours = delta.seconds // 3600
112    return f"{hours}h ago"
113 
114 
115def format_bytes(b):
116    """Format bytes as human-readable size."""
117    if not b:
118        return "—"
119    if b > 1_000_000:
120        return f"{b/1_000_000:.1f} MB"
121    if b > 1_000:
122        return f"{b/1_000:.0f} KB"
123    return f"{b} B"
124 
125 
126def show_list(client, resource_type="all"):
127    """Display current resources."""
128    if resource_type in ("all", "jobs"):
129        jobs = list_jobs(client)
130        print(f"\n📋 Fine-tuning jobs ({len(jobs)}):")
131        if jobs:
132            print(f"  {'ID':<25} {'Status':<12} {'Model':<20} {'Age':<10}")
133            print(f"  {'-'*25} {'-'*12} {'-'*20} {'-'*10}")
134            for j in jobs:
135                print(f"  {j['id'][:24]:<25} {j['status']:<12} {j['model']:<20} {format_age(j['created'])}")
136        else:
137            print("  (none)")
138 
139    if resource_type in ("all", "deployments"):
140        deps = list_deployments(client)
141        print(f"\n🚀 Fine-tuned models ({len(deps)}):")
142        if deps:
143            print(f"  {'Model':<60} {'Age':<10}")
144            print(f"  {'-'*60} {'-'*10}")
145            for d in deps:
146                name = d['model'][:59]
147                print(f"  {name:<60} {format_age(d['created'])}")
148        else:
149            print("  (none)")
150 
151    if resource_type in ("all", "files"):
152        files = list_files(client)
153        print(f"\n📁 Uploaded files ({len(files)}):")
154        if files:
155            print(f"  {'ID':<40} {'Size':>8} {'Purpose':<12} {'Age':<10} {'Status':<10}")
156            print(f"  {'-'*40} {'-'*8} {'-'*12} {'-'*10} {'-'*10}")
157            for f in files:
158                print(f"  {f['id']:<40} {format_bytes(f['bytes']):>8} {f['purpose']:<12} {format_age(f['created']):<10} {f['status']}")
159        else:
160            print("  (none)")
161 
162        # Quota warning
163        if len(files) >= 80:
164            print(f"\n  ⚠️ {len(files)}/100 file slots used — approaching quota limit!")
165 
166 
167def delete_files(client, older_than_days=None, dry_run=False):
168    """Delete uploaded files, optionally filtering by age."""
169    files = list_files(client)
170    now = datetime.now(timezone.utc)
171 
172    to_delete = []
173    for f in files:
174        if older_than_days:
175            age_days = (now - f["created"]).days
176            if age_days < older_than_days:
177                continue
178        to_delete.append(f)
179 
180    if not to_delete:
181        print("No files to delete.")
182        return
183 
184    label = f"older than {older_than_days} days" if older_than_days else "all"
185    print(f"\n{'[DRY RUN] ' if dry_run else ''}Deleting {len(to_delete)} files ({label}):")
186 
187    deleted = 0
188    for f in to_delete:
189        print(f"  {'Would delete' if dry_run else 'Deleting'}: {f['id']} ({f['filename']}, {format_age(f['created'])})")
190        if not dry_run:
191            try:
192                client.files.delete(f["id"])
193                deleted += 1
194            except Exception as e:
195                print(f"    ❌ Failed: {e}")
196 
197    if not dry_run:
198        print(f"\n✅ Deleted {deleted}/{len(to_delete)} files")
199 
200 
201def cancel_pending_jobs(client, dry_run=False):
202    """Cancel any pending or queued jobs."""
203    jobs = list_jobs(client)
204    pending = [j for j in jobs if j["status"] in ("pending", "queued", "validating_files")]
205 
206    if not pending:
207        print("No pending jobs to cancel.")
208        return
209 
210    print(f"\n{'[DRY RUN] ' if dry_run else ''}Cancelling {len(pending)} pending jobs:")
211    for j in pending:
212        print(f"  {'Would cancel' if dry_run else 'Cancelling'}: {j['id']} ({j['status']})")
213        if not dry_run:
214            try:
215                client.fine_tuning.jobs.cancel(j["id"])
216            except Exception as e:
217                print(f"    ❌ Failed: {e}")
218 
219 
220def build_parser():
221    parser = HelpOnErrorParser(
222        description="Clean up fine-tuning resources (deployments, files, jobs)",
223        epilog=(
224            "Examples:\n"
225            "  python cleanup.py --list                         # Show all resources\n"
226            "  python cleanup.py --list --type files             # Show files only\n"
227            "  python cleanup.py --delete-files --older-than 7   # Delete files older than 7 days\n"
228            "  python cleanup.py --delete-files --dry-run        # Preview what would be deleted\n"
229            "  python cleanup.py --cancel-pending                # Cancel queued jobs"
230        ),
231        formatter_class=argparse.RawTextHelpFormatter,
232    )
233    parser.add_argument("--base-url", default=os.environ.get("OPENAI_BASE_URL"), help="Project /v1/ endpoint URL")
234    parser.add_argument("--endpoint", default=os.environ.get("AZURE_OPENAI_ENDPOINT"),
235                        help="Azure OpenAI endpoint (fallback)")
236    parser.add_argument("--api-key", default=os.environ.get("AZURE_OPENAI_API_KEY"), help="API key")
237    parser.add_argument("--project-endpoint", default=os.environ.get("AZURE_AI_PROJECT_ENDPOINT"),
238                        help="Azure AI project endpoint")
239 
240    parser.add_argument("--list", action="store_true", help="List resources")
241    parser.add_argument("--type", choices=["all", "jobs", "deployments", "files"], default="all",
242                        help="Resource type to list (default: all)")
243 
244    parser.add_argument("--delete-files", action="store_true", help="Delete uploaded files")
245    parser.add_argument("--older-than", type=int, default=None,
246                        help="Only delete files older than N days (use with --delete-files)")
247    parser.add_argument("--cancel-pending", action="store_true", help="Cancel pending/queued jobs")
248    parser.add_argument("--dry-run", action="store_true", help="Preview changes without executing")
249    return parser
250 
251 
252if __name__ == "__main__":
253    parser = build_parser()
254    if len(sys.argv) == 1:
255        parser.print_help()
256        sys.exit(0)
257 
258    args = parser.parse_args()
259    client, method = get_clients(base_url=args.base_url, azure_endpoint=args.endpoint, project_endpoint=args.project_endpoint, api_key=args.api_key)
260 
261    if args.list:
262        show_list(client, args.type)
263 
264    if args.delete_files:
265        delete_files(client, older_than_days=args.older_than, dry_run=args.dry_run)
266 
267    if args.cancel_pending:
268        cancel_pending_jobs(client, dry_run=args.dry_run)
269