Update app.py
Browse files
app.py
CHANGED
|
@@ -133,24 +133,42 @@ def _log_warmup(msg: str):
|
|
| 133 |
if len(warmup_state["logs"]) > 400:
|
| 134 |
warmup_state["logs"] = warmup_state["logs"][-400:]
|
| 135 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 136 |
def _download_one(repo_id: str, tries: int = 3) -> bool:
|
| 137 |
"""
|
| 138 |
-
Télécharge un repo HF en réessayant si besoin.
|
| 139 |
"""
|
| 140 |
cache_home = os.path.expanduser(os.getenv("HF_HOME", "/home/user/.cache/huggingface"))
|
| 141 |
local_dir = os.path.join(cache_home, "models", repo_id.replace("/", "__"))
|
| 142 |
-
|
|
|
|
| 143 |
try:
|
| 144 |
if _is_repo_cached(repo_id):
|
| 145 |
-
|
|
|
|
| 146 |
return True
|
| 147 |
except Exception:
|
| 148 |
-
# En cas d'erreur d'audit, on continue le process normal
|
| 149 |
pass
|
| 150 |
|
| 151 |
for attempt in range(1, tries + 1):
|
| 152 |
if warmup_stop.is_set():
|
|
|
|
| 153 |
return False
|
|
|
|
|
|
|
|
|
|
| 154 |
try:
|
| 155 |
snapshot_download(
|
| 156 |
repo_id,
|
|
@@ -158,9 +176,13 @@ def _download_one(repo_id: str, tries: int = 3) -> bool:
|
|
| 158 |
local_dir_use_symlinks=False,
|
| 159 |
resume_download=True,
|
| 160 |
)
|
|
|
|
|
|
|
|
|
|
| 161 |
return True
|
| 162 |
except Exception as e:
|
| 163 |
-
|
|
|
|
| 164 |
time.sleep(min(10, 2 * attempt))
|
| 165 |
return False
|
| 166 |
|
|
@@ -178,33 +200,39 @@ def _warmup_thread(models: List[str]):
|
|
| 178 |
})
|
| 179 |
warmup_stop.clear()
|
| 180 |
ok_count = 0
|
|
|
|
|
|
|
|
|
|
| 181 |
for i, repo in enumerate(models):
|
| 182 |
if warmup_stop.is_set():
|
| 183 |
-
_log_warmup("Arrêt demandé par l’utilisateur
|
| 184 |
break
|
|
|
|
| 185 |
with warmup_lock:
|
| 186 |
warmup_state["idx"] = i
|
| 187 |
warmup_state["current"] = repo
|
| 188 |
warmup_state["percent"] = int((i / max(1, len(models))) * 100)
|
| 189 |
-
|
| 190 |
ok = _download_one(repo)
|
| 191 |
if ok:
|
| 192 |
ok_count += 1
|
| 193 |
-
_log_warmup(f"OK: {repo}")
|
| 194 |
with warmup_lock:
|
| 195 |
warmup_state["ok_count"] = ok_count
|
| 196 |
-
|
| 197 |
-
_log_warmup(f"ÉCHEC: {repo}")
|
| 198 |
-
# Met à jour la progression globale après ce repo (ok ou échec)
|
| 199 |
with warmup_lock:
|
| 200 |
warmup_state["percent"] = int(((i + 1) / max(1, len(models))) * 100)
|
| 201 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 202 |
with warmup_lock:
|
| 203 |
warmup_state["percent"] = 100
|
| 204 |
warmup_state["done"] = True
|
| 205 |
warmup_state["running"] = False
|
| 206 |
warmup_state["ok_count"] = ok_count
|
| 207 |
-
_log_warmup(f"
|
|
|
|
| 208 |
# <<<< HUGINFACE PATCH: WARMUP STATE+HELPERS END >>>>
|
| 209 |
# ---------- Helpers ----------
|
| 210 |
|
|
|
|
| 133 |
if len(warmup_state["logs"]) > 400:
|
| 134 |
warmup_state["logs"] = warmup_state["logs"][-400:]
|
| 135 |
|
| 136 |
+
def _dir_size_bytes(path: str) -> int:
|
| 137 |
+
try:
|
| 138 |
+
total = 0
|
| 139 |
+
for root, _, files in os.walk(path):
|
| 140 |
+
for f in files:
|
| 141 |
+
try:
|
| 142 |
+
total += os.path.getsize(os.path.join(root, f))
|
| 143 |
+
except Exception:
|
| 144 |
+
pass
|
| 145 |
+
return total
|
| 146 |
+
except Exception:
|
| 147 |
+
return 0
|
| 148 |
+
|
| 149 |
def _download_one(repo_id: str, tries: int = 3) -> bool:
|
| 150 |
"""
|
| 151 |
+
Télécharge un repo HF en réessayant si besoin, avec logs précis.
|
| 152 |
"""
|
| 153 |
cache_home = os.path.expanduser(os.getenv("HF_HOME", "/home/user/.cache/huggingface"))
|
| 154 |
local_dir = os.path.join(cache_home, "models", repo_id.replace("/", "__"))
|
| 155 |
+
|
| 156 |
+
# Cache déjà présent ?
|
| 157 |
try:
|
| 158 |
if _is_repo_cached(repo_id):
|
| 159 |
+
sz = _dir_size_bytes(local_dir)
|
| 160 |
+
_log_warmup(f"[CACHE] {repo_id} • {local_dir} • {sz/1e6:.1f} MB (skip)")
|
| 161 |
return True
|
| 162 |
except Exception:
|
|
|
|
| 163 |
pass
|
| 164 |
|
| 165 |
for attempt in range(1, tries + 1):
|
| 166 |
if warmup_stop.is_set():
|
| 167 |
+
_log_warmup(f"[STOP] Abandon demandé avant téléchargement de {repo_id}")
|
| 168 |
return False
|
| 169 |
+
|
| 170 |
+
t0 = time.time()
|
| 171 |
+
_log_warmup(f"[START] {repo_id} • tentative {attempt}/{tries} • {local_dir}")
|
| 172 |
try:
|
| 173 |
snapshot_download(
|
| 174 |
repo_id,
|
|
|
|
| 176 |
local_dir_use_symlinks=False,
|
| 177 |
resume_download=True,
|
| 178 |
)
|
| 179 |
+
dt = time.time() - t0
|
| 180 |
+
sz = _dir_size_bytes(local_dir)
|
| 181 |
+
_log_warmup(f"[DONE] {repo_id} • {dt:.1f}s • {sz/1e6:.1f} MB")
|
| 182 |
return True
|
| 183 |
except Exception as e:
|
| 184 |
+
dt = time.time() - t0
|
| 185 |
+
_log_warmup(f"[FAIL] {repo_id} • {dt:.1f}s • {type(e).__name__}: {e}")
|
| 186 |
time.sleep(min(10, 2 * attempt))
|
| 187 |
return False
|
| 188 |
|
|
|
|
| 200 |
})
|
| 201 |
warmup_stop.clear()
|
| 202 |
ok_count = 0
|
| 203 |
+
t_global = time.time()
|
| 204 |
+
_log_warmup(f"[JOB] start • {len(models)} dépôts")
|
| 205 |
+
|
| 206 |
for i, repo in enumerate(models):
|
| 207 |
if warmup_stop.is_set():
|
| 208 |
+
_log_warmup("[STOP] Arrêt demandé par l’utilisateur — fin du job après ce point")
|
| 209 |
break
|
| 210 |
+
|
| 211 |
with warmup_lock:
|
| 212 |
warmup_state["idx"] = i
|
| 213 |
warmup_state["current"] = repo
|
| 214 |
warmup_state["percent"] = int((i / max(1, len(models))) * 100)
|
| 215 |
+
|
| 216 |
ok = _download_one(repo)
|
| 217 |
if ok:
|
| 218 |
ok_count += 1
|
|
|
|
| 219 |
with warmup_lock:
|
| 220 |
warmup_state["ok_count"] = ok_count
|
| 221 |
+
# progression après dépôt (OK ou FAIL)
|
|
|
|
|
|
|
| 222 |
with warmup_lock:
|
| 223 |
warmup_state["percent"] = int(((i + 1) / max(1, len(models))) * 100)
|
| 224 |
|
| 225 |
+
if warmup_stop.is_set():
|
| 226 |
+
_log_warmup("[STOP] Fin anticipée — demande reçue pendant le run")
|
| 227 |
+
break
|
| 228 |
+
|
| 229 |
with warmup_lock:
|
| 230 |
warmup_state["percent"] = 100
|
| 231 |
warmup_state["done"] = True
|
| 232 |
warmup_state["running"] = False
|
| 233 |
warmup_state["ok_count"] = ok_count
|
| 234 |
+
_log_warmup(f"[JOB] done • {ok_count}/{len(models)} • {time.time()-t_global:.1f}s")
|
| 235 |
+
|
| 236 |
# <<<< HUGINFACE PATCH: WARMUP STATE+HELPERS END >>>>
|
| 237 |
# ---------- Helpers ----------
|
| 238 |
|