mirror of
https://github.com/R0m1k3/Loki.git
synced 2026-10-17 04:10:04 +02:00
Claude
2da2efc1fd
La façade sert /api/start, /api/stop, /api/restart, /api/vram/unload et /api/vram/reload quand LOKI_ENGINE=rust (expérimental ; le Go reste le défaut). Elle lance elle-même « loki-rs engine <action> », le superviseur que le Go appelait déjà en conteneur, avec l'environnement qu'il lui passait et de quoi joindre l'API interne du Go ; sa dernière ligne d'erreur devient l'erreur rendue. Le contrôle d'avant démarrage (preflightEngine : preset externe, Strata, BIN, MODEL, tranches) et le bilan de VRAM du déchargement (gpuSettle) sont portés. La façade ne tient ce lot qu'en conteneur : sous systemd ou avec LOKI_ENGINE_SUPERVISOR=go, loki-rs ne passe pas la variable au Go. Restent au Go derrière son API interne (facade_engine.go) : le verrou de l'optimiseur avant tout démarrage (/internal/v1/service/prestart), la réponse en cours quand il tient la conversation (/internal/v1/chat/generating) et l'arrêt de la dictée quand il la tient (/internal/v1/dictate/stop). Parité : parity_engine_test.go tire du Go dix bilans de VRAM, vingt-six installations passées au préflight, les messages du superviseur et dix réponses de route, rejoués par tests/enginectl.rs ; démarrage, arrêt, redémarrage, déchargement chiffré et rechargement vérifiés de bout en bout avec un faux superviseur et un faux nvidia-smi ; TestFacadeMoteur côté Go. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016tcGfJWefW5kUGoC2wL8cv
552 lines
12 KiB
JSON
552 lines
12 KiB
JSON
{
|
|
"settle": [
|
|
{
|
|
"before": 14000,
|
|
"out": 500,
|
|
"seq": [
|
|
20000,
|
|
20000,
|
|
12000,
|
|
3000,
|
|
500,
|
|
500
|
|
]
|
|
},
|
|
{
|
|
"before": 14000,
|
|
"out": 400,
|
|
"seq": [
|
|
15000,
|
|
8000,
|
|
400,
|
|
400
|
|
]
|
|
},
|
|
{
|
|
"before": 14000,
|
|
"out": 600,
|
|
"seq": [
|
|
14000,
|
|
-1,
|
|
600,
|
|
600
|
|
]
|
|
},
|
|
{
|
|
"before": 14000,
|
|
"out": 14000,
|
|
"seq": [
|
|
-1,
|
|
-1,
|
|
-1,
|
|
-1,
|
|
-1,
|
|
-1,
|
|
-1,
|
|
-1
|
|
]
|
|
},
|
|
{
|
|
"before": 14000,
|
|
"out": 14000,
|
|
"seq": [
|
|
14000,
|
|
14000,
|
|
14000,
|
|
14000,
|
|
14000,
|
|
14000,
|
|
14000,
|
|
14000
|
|
]
|
|
},
|
|
{
|
|
"before": 0,
|
|
"out": 0,
|
|
"seq": [
|
|
0,
|
|
0,
|
|
0
|
|
]
|
|
},
|
|
{
|
|
"before": 9000,
|
|
"out": 8999,
|
|
"seq": [
|
|
9000,
|
|
8999,
|
|
8999
|
|
]
|
|
},
|
|
{
|
|
"before": 9000,
|
|
"out": 100,
|
|
"seq": [
|
|
-1,
|
|
9000,
|
|
-1,
|
|
9000,
|
|
100,
|
|
-1,
|
|
100
|
|
]
|
|
},
|
|
{
|
|
"before": 5000,
|
|
"out": 7000,
|
|
"seq": [
|
|
6000,
|
|
7000,
|
|
7000,
|
|
7000,
|
|
7000,
|
|
7000,
|
|
7000,
|
|
7000
|
|
]
|
|
},
|
|
{
|
|
"before": 5000,
|
|
"out": 600,
|
|
"seq": [
|
|
4000,
|
|
3000,
|
|
2000,
|
|
1000,
|
|
900,
|
|
800,
|
|
700,
|
|
600
|
|
]
|
|
}
|
|
],
|
|
"last_error": {
|
|
"": "",
|
|
"[loki-rs engine] service loki-engine non démarré\n": "service loki-engine non démarré",
|
|
"[ok] démarré\n[attention] 2 processus du moteur toujours là\n": "[attention] 2 processus du moteur toujours là",
|
|
"ligne 1\n[loki-rs engine] moteur absent \n\n": "moteur absent",
|
|
"pas de préfixe": "pas de préfixe"
|
|
},
|
|
"plain": {
|
|
"\u001b[1;36mcouleur\u001b[0m et \u001b[32mvert\u001b[0m": "couleur et vert",
|
|
"\u001b[mvide\u001b[0;1;2m": "vide",
|
|
"BIN non défini — installe un moteur : \u001b[1mloki llamacpp install\u001b[0m": "BIN non défini — installe un moteur : loki llamacpp install",
|
|
"rien à retirer": "rien à retirer"
|
|
},
|
|
"preflight": [
|
|
{
|
|
"name": "externe",
|
|
"config": {
|
|
"EXTERNAL": "1"
|
|
},
|
|
"out": ""
|
|
},
|
|
{
|
|
"name": "externe sans BIN ni MODEL",
|
|
"config": {
|
|
"BIN": "",
|
|
"EXTERNAL": " 1 "
|
|
},
|
|
"out": ""
|
|
},
|
|
{
|
|
"name": "strata sans config",
|
|
"config": {
|
|
"ENGINE": "strata"
|
|
},
|
|
"out": "STRATA_CONFIG non défini : réinstalle le modèle depuis Paramètres → Moteur → Strata"
|
|
},
|
|
{
|
|
"name": "strata config absente",
|
|
"config": {
|
|
"ENGINE": "Strata",
|
|
"STRATA_CONFIG": "{{HOME}}/strata/m/config.json"
|
|
},
|
|
"out": "configuration de Strata introuvable : {{HOME}}/strata/m/config.json — réinstalle le modèle"
|
|
},
|
|
{
|
|
"name": "strata sans venv",
|
|
"config": {
|
|
"ENGINE": "strata",
|
|
"STRATA_CONFIG": "{{HOME}}/strata/m/config.json"
|
|
},
|
|
"files": [
|
|
"strata/m/config.json"
|
|
],
|
|
"out": "environnement Python de Strata absent ({{HOME}}/strata/m/.venv/bin/python) : réinstalle le modèle"
|
|
},
|
|
{
|
|
"name": "strata prêt",
|
|
"config": {
|
|
"ENGINE": "moe",
|
|
"STRATA_CONFIG": "{{HOME}}/strata/m/config.json"
|
|
},
|
|
"files": [
|
|
"strata/m/config.json",
|
|
"strata/m/.venv/bin/python"
|
|
],
|
|
"out": ""
|
|
},
|
|
{
|
|
"name": "BIN vide",
|
|
"config": {
|
|
"MODEL": "m.gguf"
|
|
},
|
|
"out": "BIN non défini — installe un moteur : loki llamacpp install (ou renseigne BIN avec loki edit)"
|
|
},
|
|
{
|
|
"name": "BIN espaces",
|
|
"config": {
|
|
"BIN": " ",
|
|
"MODEL": "m.gguf"
|
|
},
|
|
"out": "BIN non défini — installe un moteur : loki llamacpp install (ou renseigne BIN avec loki edit)"
|
|
},
|
|
{
|
|
"name": "BIN relatif absent",
|
|
"config": {
|
|
"BIN": "bin/absent",
|
|
"MODEL": "m.gguf"
|
|
},
|
|
"out": "moteur introuvable : {{HOME}}/bin/absent — relance loki llamacpp install"
|
|
},
|
|
{
|
|
"name": "BIN absolu absent",
|
|
"config": {
|
|
"BIN": "{{HOME}}/ailleurs/llama-server",
|
|
"MODEL": "m.gguf"
|
|
},
|
|
"out": "moteur introuvable : {{HOME}}/ailleurs/llama-server — relance loki llamacpp install"
|
|
},
|
|
{
|
|
"name": "BIN dossier",
|
|
"config": {
|
|
"BIN": "bin/",
|
|
"MODEL": "m.gguf"
|
|
},
|
|
"files": [
|
|
"bin/"
|
|
],
|
|
"out": "modèle introuvable : {{HOME}}/models/m.gguf — corrige MODEL avec loki edit"
|
|
},
|
|
{
|
|
"name": "MODEL vide",
|
|
"config": {
|
|
"BIN": "bin/llama-server"
|
|
},
|
|
"files": [
|
|
"bin/llama-server"
|
|
],
|
|
"out": "MODEL non défini — indique un .gguf avec loki edit (dossier des modèles : {{HOME}}/models)"
|
|
},
|
|
{
|
|
"name": "MODEL absent",
|
|
"config": {
|
|
"BIN": "bin/llama-server",
|
|
"MODEL": "absent.gguf"
|
|
},
|
|
"files": [
|
|
"bin/llama-server"
|
|
],
|
|
"out": "modèle introuvable : {{HOME}}/models/absent.gguf — corrige MODEL avec loki edit"
|
|
},
|
|
{
|
|
"name": "MODEL relatif",
|
|
"config": {
|
|
"BIN": "{{HOME}}/bin/llama-server",
|
|
"MODEL": "m.gguf"
|
|
},
|
|
"files": [
|
|
"bin/llama-server",
|
|
"models/m.gguf"
|
|
],
|
|
"out": ""
|
|
},
|
|
{
|
|
"name": "MODEL remonte",
|
|
"config": {
|
|
"BIN": "bin/llama-server",
|
|
"MODEL": "../m.gguf"
|
|
},
|
|
"files": [
|
|
"bin/llama-server",
|
|
"models/m.gguf"
|
|
],
|
|
"out": ""
|
|
},
|
|
{
|
|
"name": "MODEL sous-dossier",
|
|
"config": {
|
|
"BIN": "bin/llama-server",
|
|
"MODEL": "sub/m.gguf"
|
|
},
|
|
"files": [
|
|
"bin/llama-server",
|
|
"models/m.gguf"
|
|
],
|
|
"out": ""
|
|
},
|
|
{
|
|
"name": "MODEL absolu hors dossiers",
|
|
"config": {
|
|
"BIN": "bin/llama-server",
|
|
"MODEL": "{{HOME}}/ailleurs/x.gguf"
|
|
},
|
|
"files": [
|
|
"bin/llama-server",
|
|
"ailleurs/x.gguf"
|
|
],
|
|
"out": ""
|
|
},
|
|
{
|
|
"name": "MODEL absolu pas gguf",
|
|
"config": {
|
|
"BIN": "bin/llama-server",
|
|
"MODEL": "{{HOME}}/ailleurs/x.bin"
|
|
},
|
|
"files": [
|
|
"bin/llama-server",
|
|
"ailleurs/x.bin"
|
|
],
|
|
"out": "MODEL={{HOME}}/ailleurs/x.bin : le modèle doit être un fichier .gguf"
|
|
},
|
|
{
|
|
"name": "MODEL absolu absent",
|
|
"config": {
|
|
"BIN": "bin/llama-server",
|
|
"MODEL": "{{HOME}}/ailleurs/y.gguf"
|
|
},
|
|
"files": [
|
|
"bin/llama-server"
|
|
],
|
|
"out": "MODEL={{HOME}}/ailleurs/y.gguf : dossier non autorisé : {{HOME}}/ailleurs — ajoute-le dans « Dossiers de modèles »"
|
|
},
|
|
{
|
|
"name": "MODEL dossier déclaré",
|
|
"config": {
|
|
"BIN": "bin/llama-server",
|
|
"MODEL": "d.gguf"
|
|
},
|
|
"files": [
|
|
"bin/llama-server",
|
|
"disque/d.gguf"
|
|
],
|
|
"dirs": [
|
|
"disque"
|
|
],
|
|
"out": ""
|
|
},
|
|
{
|
|
"name": "MODEL chemin dans dossier déclaré",
|
|
"config": {
|
|
"BIN": "bin/llama-server",
|
|
"MODEL": "{{HOME}}/disque/d.gguf"
|
|
},
|
|
"files": [
|
|
"bin/llama-server",
|
|
"disque/d.gguf"
|
|
],
|
|
"dirs": [
|
|
"disque"
|
|
],
|
|
"out": ""
|
|
},
|
|
{
|
|
"name": "tranches manquantes",
|
|
"config": {
|
|
"BIN": "bin/llama-server",
|
|
"MODEL": "big-00001-of-00003.gguf"
|
|
},
|
|
"files": [
|
|
"bin/llama-server",
|
|
"models/big-00001-of-00003.gguf",
|
|
"models/big-00003-of-00003.gguf"
|
|
],
|
|
"out": "modèle incomplet : il manque big-00002-of-00003.gguf dans {{HOME}}/models — ce modèle tient en 3 fichiers, télécharge-les tous"
|
|
},
|
|
{
|
|
"name": "tranches dossier",
|
|
"config": {
|
|
"BIN": "bin/llama-server",
|
|
"MODEL": "big-00001-of-00002.gguf"
|
|
},
|
|
"files": [
|
|
"bin/llama-server",
|
|
"models/big-00001-of-00002.gguf",
|
|
"models/big-00002-of-00002.gguf/"
|
|
],
|
|
"out": "modèle incomplet : il manque big-00002-of-00002.gguf dans {{HOME}}/models — ce modèle tient en 2 fichiers, télécharge-les tous"
|
|
},
|
|
{
|
|
"name": "tranches complètes",
|
|
"config": {
|
|
"BIN": "bin/llama-server",
|
|
"MODEL": "big-00001-of-00002.gguf"
|
|
},
|
|
"files": [
|
|
"bin/llama-server",
|
|
"models/big-00001-of-00002.gguf",
|
|
"models/big-00002-of-00002.gguf"
|
|
],
|
|
"out": ""
|
|
},
|
|
{
|
|
"name": "tranche 2 désignée",
|
|
"config": {
|
|
"BIN": "bin/llama-server",
|
|
"MODEL": "big-00002-of-00002.gguf"
|
|
},
|
|
"files": [
|
|
"bin/llama-server",
|
|
"models/big-00002-of-00002.gguf"
|
|
],
|
|
"out": "modèle incomplet : il manque big-00001-of-00002.gguf dans {{HOME}}/models — ce modèle tient en 2 fichiers, télécharge-les tous"
|
|
},
|
|
{
|
|
"name": "MODEL guillemets",
|
|
"config": {
|
|
"BIN": "bin/llama-server",
|
|
"MODEL": " m.gguf "
|
|
},
|
|
"files": [
|
|
"bin/llama-server",
|
|
"models/m.gguf"
|
|
],
|
|
"out": ""
|
|
}
|
|
],
|
|
"steps": [
|
|
{
|
|
"name": "déchargement en GET",
|
|
"method": "GET",
|
|
"path": "/api/vram/unload",
|
|
"status": 405,
|
|
"allow": "POST",
|
|
"out": {
|
|
"error": "méthode GET refusée : POST attendu",
|
|
"ok": false
|
|
}
|
|
},
|
|
{
|
|
"name": "rechargement en GET",
|
|
"method": "GET",
|
|
"path": "/api/vram/reload",
|
|
"status": 405,
|
|
"allow": "POST",
|
|
"out": {
|
|
"error": "méthode GET refusée : POST attendu",
|
|
"ok": false
|
|
}
|
|
},
|
|
{
|
|
"name": "rechargement en PUT",
|
|
"method": "PUT",
|
|
"path": "/api/vram/reload",
|
|
"status": 405,
|
|
"allow": "POST",
|
|
"out": {
|
|
"error": "méthode PUT refusée : POST attendu",
|
|
"ok": false
|
|
}
|
|
},
|
|
{
|
|
"name": "génération en cours",
|
|
"method": "POST",
|
|
"path": "/api/vram/unload",
|
|
"body": "{}",
|
|
"generating": true,
|
|
"status": 409,
|
|
"out": {
|
|
"error": "une réponse est en cours d'écriture — décharger maintenant l'interromprait",
|
|
"generating": true,
|
|
"ok": false
|
|
}
|
|
},
|
|
{
|
|
"name": "génération en cours, corps illisible",
|
|
"method": "POST",
|
|
"path": "/api/vram/unload",
|
|
"body": "nimp",
|
|
"generating": true,
|
|
"status": 409,
|
|
"out": {
|
|
"error": "une réponse est en cours d'écriture — décharger maintenant l'interromprait",
|
|
"generating": true,
|
|
"ok": false
|
|
}
|
|
},
|
|
{
|
|
"name": "moteur déjà arrêté",
|
|
"method": "POST",
|
|
"path": "/api/vram/unload",
|
|
"body": "{\"force\":true}",
|
|
"generating": true,
|
|
"status": 200,
|
|
"out": {
|
|
"active": false,
|
|
"before_mb": 0,
|
|
"error": "",
|
|
"freed_mb": 0,
|
|
"gpu": false,
|
|
"ok": true,
|
|
"used_mb": 0,
|
|
"was_active": false
|
|
}
|
|
},
|
|
{
|
|
"name": "moteur arrêté, sans corps",
|
|
"method": "POST",
|
|
"path": "/api/vram/unload",
|
|
"status": 200,
|
|
"out": {
|
|
"active": false,
|
|
"before_mb": 0,
|
|
"error": "",
|
|
"freed_mb": 0,
|
|
"gpu": false,
|
|
"ok": true,
|
|
"used_mb": 0,
|
|
"was_active": false
|
|
}
|
|
},
|
|
{
|
|
"name": "optimisation en cours",
|
|
"method": "POST",
|
|
"path": "/api/vram/reload",
|
|
"tune": true,
|
|
"status": 409,
|
|
"out": {
|
|
"error": "optimisation en cours (loki tune) : le moteur est réservé aux essais — attends la fin ou annule-la",
|
|
"ok": false,
|
|
"tuning": true
|
|
}
|
|
},
|
|
{
|
|
"name": "préflight sans BIN",
|
|
"method": "POST",
|
|
"path": "/api/vram/reload",
|
|
"config": {
|
|
"MODEL": "m.gguf"
|
|
},
|
|
"status": 200,
|
|
"out": {
|
|
"error": "BIN non défini — installe un moteur : loki llamacpp install (ou renseigne BIN avec loki edit)",
|
|
"ok": false
|
|
}
|
|
},
|
|
{
|
|
"name": "préflight sans MODEL",
|
|
"method": "POST",
|
|
"path": "/api/vram/reload",
|
|
"config": {
|
|
"BIN": "/bin/sh"
|
|
},
|
|
"status": 200,
|
|
"out": {
|
|
"error": "MODEL non défini — indique un .gguf avec loki edit (dossier des modèles : {{HOME}}/models)",
|
|
"ok": false
|
|
}
|
|
}
|
|
]
|
|
}
|