Bienvenue sur le site admin-sys

exemple de configuration opencode

Jerome ROBERT — 2026-06-09T10:56:46Z

/.opencode/config$ cat opencode.json <== Devstral-Small-2-24B-Instruct-2512-UD-Q8_K_XL.gguf

 { "$schema": "https://opencode.ai/config.json", "model": "local/Devstral", "provider": { "local": { "name": "Local Llama.cpp Server", "npm": "@ai-sdk/openai", "options": { "baseURL": "http://127.0.0.1:9090/v1", "apiKey": "none" }, "models": { "Devstral": { "id": "Devstral-Small-2-24B-Instruct-2512-UD-Q8_K_XL.gguf", // ID exact du modèle (comme dans la réponse curl) "name": "Devstral-Small-2-24B-Instruct" } } } } }

/.opencode/config$ cat opencode.json <== Qwen3.6-35B-A3B-Q8_0}

  { "$schema": "https://opencode.ai/config.json", "model": "local/qwen35b", "provider": { "local": { "name": "Local Llama.cpp Server", "npm": "@ai-sdk/openai", "options": { "baseURL": "http://127.0.0.1:9090/v1" }, "models": { "qwen35b": { "id": "qwen35b", "name": "Qwen3.6-35B-A3B-Q8_0" } } } } }

le modele est lancer par : run-qwen36-q8-V2.sh

 # !/bin/bash
 MODEL=./models/Qwen3.6-35B-A3B-Q8_0.gguf
 MODEL_DIR="/home/jerome/models"
 LLAMA_BIN="/home/jerome/llama.cpp/build/bin/llama-server"
 # ===========================
 # CONFIG GPU (ROCm)
 # ===========================
 export HIP_VISIBLE_DEVICES=0,1
 export ROCM_VISIBLE_DEVICES=0,1
 # ===========================
 # LANCEMENT LLAMA.CPP
 # ===========================
 $LLAMA_BIN
 -m $MODEL
 —ctx-size 158000
 —n-gpu-layers 999
 —tensor-split 1,1
 —cache-type-k q8_0
 —cache-type-v q8_0
 —port 9090
 —host 127.0.0.1
 —jinja
 —reasoning off
 —temp 0.3
 —top-k 20
 —top-p 0.8
 —repeat-penalty 1.05

le modele Devstral-Small-2-24B-Instruct-2512-UD-Q8_K_XL.gguf est lancer par : run-devmistral-q8.sh

# !/bin/bash
MODEL=./models/Devstral-Small-2-24B-Instruct-2512-UD-Q8_K_XL.gguf
MODEL_DIR="/home/jerome/models"
LLAMA_BIN="/home/jerome/llama.cpp/build/bin/llama-server"

# ===========================
# CONFIG GPU (ROCm)
# ===========================
export HIP_VISIBLE_DEVICES=0,1
export ROCM_VISIBLE_DEVICES=0,1

# ===========================
# LANCEMENT LLAMA.CPP
# ===========================
$LLAMA_BIN
-m $MODEL
—ctx-size 158000
—n-gpu-layers 999
—tensor-split 1,1
—cache-type-k q8_0
—cache-type-v q8_0
—port 9090
—host 127.0.0.1
—jinja
—reasoning off
—temp 0.2
—top-k 20
—top-p 0.8
—repeat-penalty 1.05

Exemple de scripts de lancement de modele :

Jerome ROBERT — 2026-05-23T11:41:12Z

Voici des exemples de scripts de lancement de modèle dans un environnement ROCm.
( Ces derniers ne sont pas "optimiser" mais cela peut aider. )

run-qwen36-q8-V2.sh

 # !/bin/bash MODEL=./models/Qwen3.6-35B-A3B-Q8_0.gguf MODEL_DIR="/home/jerome/models" LLAMA_BIN="/home/jerome/llama.cpp/build/bin/llama-server" # =========================== # CONFIG GPU (ROCm) # =========================== export HIP_VISIBLE_DEVICES=0,1 export ROCM_VISIBLE_DEVICES=0,1 # =========================== # LANCEMENT LLAMA.CPP # =========================== $LLAMA_BIN \ -m $MODEL \ —ctx-size 158000 \ —n-gpu-layers 999 \ —tensor-split 1,1 \ —cache-type-k q8_0 \ —cache-type-v q8_0 \ —port 9090 \ —host 127.0.0.1 \ —jinja \ —chat-template-kwargs '"enable_thinking" : false' \ —temp 0.3 \ —top-k 20 \ —top-p 0.8 \ —repeat-penalty 1.05

lanceDocker-2-GPU-deepseek-v2-lite_code.sh

 # !/bin/bash export TORCHDYNAMO_DISABLE=1 export VLLM_USE_TRITON_FLASH_ATTN=0 docker run —rm -it —device=/dev/kfd —device=/dev/dri —group-add video —shm-size=16g —ipc=host -v /.cache/huggingface :/root/.cache/huggingface -p 8000:8000 -e CUDA_VISIBLE_DEVICES=0,1 -e VLLM_ROCM_USE_AITER=0 rocm/vllm-dev:rocm7.1.1_navi_ubuntu24.04_py3.12_pytorch_2.8_vllm_0.10.2rc1 vllm serve deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct —served-model-name deepseek-v2-lite —tensor-parallel-size 2 —dtype float16 —gpu-memory-utilization 0.85 —max-model-len 32768 —max-num-seqs 16 —enforce-eager

lanceDocker-deepseek-coder-1GPU.sh

 docker run —rm -d —name vllm-deepseek-coder —device=/dev/kfd —device=/dev/dri —group-add video -e HIP_VISIBLE_DEVICES=1 -e HSA_OVERRIDE_GFX_VERSION=11.0.0 -e VLLM_USE_MODELSCOPE=false —shm-size=16g —ipc=host -v /.cache/huggingface :/root/.cache/huggingface -p 5002:8000 rocm/vllm-dev:rocm7.1.1_navi_ubuntu24.04_py3.12_pytorch_2.8_vllm_0.10.2rc1 vllm serve deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct —gpu-memory-utilization 0.65 —max-model-len 8192 —served-model-name deepseek-coder-v2-lite —enforce-eager —disable-log-requests —host 0.0.0.0 —port 8000

lanceDocker-deepseek-coder-ok-2GPU.sh

 docker run —rm -d —name vllm-deepseek-coder —device=/dev/kfd —device=/dev/dri —group-add video —shm-size=16g —ipc=host -v /.cache/huggingface :/root/.cache/huggingface -p 8002:8000 rocm/vllm-dev:rocm7.1.1_navi_ubuntu24.04_py3.12_pytorch_2.8_vllm_0.10.2rc1 vllm serve deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct —tensor-parallel-size 2 —gpu-memory-utilization 0.85 —max-model-len 16384 —served-model-name deepseek-coder-v2-lite —host 0.0.0.0 —port 8000 echo "Logs : docker logs -f vllm-deepseek-coder" echo "Stop : docker stop vllm-deepseek-coder"

lance-modele-pour-opencodeV2.sh

 # !/bin/bash # ============================ # START LLAMA DUAL SERVERS # ============================ MODEL_DIR="/home/jerome/models" LLAMA_BIN="/home/jerome/llama.cpp/build/bin/llama-server" PLAN_MODEL="$MODEL_DIR/Qwen2.5-14B-Instruct-Q8_0.gguf" BUILD_MODEL="$MODEL_DIR/qwen2.5-coder-14b-instruct-q8_0.gguf" LOG_PLAN="/home/jerome/models/plan.log" LOG_BUILD="/home/jerome/models/build.log" echo "=== Lancement des deux modèles avec contexte optimisé ===" ############################################ # 1) PLAN sur GPU0 ############################################ echo "[PLAN] GPU0 ? $PLAN_MODEL" export HIP_VISIBLE_DEVICES=0 "$LLAMA_BIN" -m "$PLAN_MODEL" —split-mode none —no-mmap —ctx-size 100000 —port 5001 —host 0.0.0.0 > "$LOG_PLAN" 2>&1 & sleep 2 ############################################ # 2) BUILD sur GPU1 ############################################ echo "[BUILD] GPU1 ? $BUILD_MODEL" export HIP_VISIBLE_DEVICES=1 docker run —rm -d —name vllm-deepseek-coder —device=/dev/kfd —device=/dev/dri —group-add video -e HIP_VISIBLE_DEVICES=1 -e HSA_OVERRIDE_GFX_VERSION=11.0.0 —shm-size=16g —ipc=host -v /.cache/huggingface :/root/.cache/huggingface -p 5002:8000 rocm/vllm-dev:rocm7.1.1_navi_ubuntu24.04_py3.12_pytorch_2.8_vllm_0.10.2rc1 vllm serve deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct —gpu-memory-utilization 0.80 —max-model-len 16384 —served-model-name deepseek-coder-v2-lite —host 0.0.0.0 —port 8000 > "$LOG_BUILD" 2>&1 & sleep 2 echo "==========================================================" echo " SERVEURS LANCÉS AVEC CONTEXTE 100k TOKENS" echo " PLAN ? http://localhost:5001" echo " BUILD ? http://localhost:5002" echo "=========================================================="

Matériel et Environnement

Jerome ROBERT — 2026-05-21T07:03:10Z

Pour réaliser mon serveur IA, j'ai choisi :

les GPU : AMD Radeon™ AI PRO R9700 32Go (Pour le prix d'une RTX 5090, j'ai pu avoir 2 GPU R9700).
carte mère WRX80 creator R2.0
128 Go RAM (32 Go peuvent suffir si tout le modele est sur les GPU
AMD Ryzen Threadripper PRO 3945WX (car le moins chère)
un boitier tour Fractal Define 7XL (vraiment super !!)
disque dur 3.5 pour proxmox
Disque Crucial P310 pour les VM (suffissent)
une "petite" alimentation de 1600W. (possible de mettre des alimentations externe.)

Point important, les GPU AMD n'ont pas de cœurs CUDA mais il y a ROCm !!!!

ROCm permet la création d'une VRAM Virtuel donc avec 2 GPU de 32 Go ==> cela fait 64 Go VRAM !!!
contre parti, plus de latence car temps de synchronisation

Ayant des panneaux solaire, c'est une IA solaire ;)

Commandes utiles

Jerome ROBERT — 2023-09-26T09:12:30Z

Commandes utiles

Hypothèse :

le nom de la baie netapp sera : filer0

Liste des volumes :

 ssh root@filer0 vol status | egrep 'online|offline|restrict' | awk '{ print $1 }' |sort

Centreon : Supervision apache

Jerome ROBERT — 2023-09-19T14:30:30Z

Préparation de la configuration apache :

Dans notre exemple le serveur WEB a pour nom : WEB001 / 192.168.141.40

AuthUserFile /var/www/.htpasswd_status
AuthName "Identifiez vous :"
AuthType Basic
Require valid-user

SetHandler server-status
Order deny,allow
Deny from all
Allow from 192.168.1.2 10.0.2.3 #à modifier ;)

htpasswd -c /var/www/.htpassword_status supervision
New password : xxxxxxxx
Re-type new password : xxxxxxxx
Adding password for user supervision

Plugin centreon

Le répertoire /var/lib/centreon/centplugins/ contiendra

# ./centreon_plugins.pl plugin=apps::apache::serverstatus::plugin list-mode

Plugin Description :
Check Apache and Apache 2 Web Servers through server-status webpage

Global Options :
* mode Choose a mode.

* dyn-mode
Specify a mode with the path (separated by ' : :').

* list-mode
List available modes.

* mode-version
Check minimal version of mode. If not, unknown error.

* version
Display plugin version.

Modes Available :
cpuload
requests
responsetime
slotstates
workers

Exemple de commande et résultat :

./centreon_plugins.pl* plugin=apps::apache::serverstatus::plugin* mode=slotstates* hostname=192.168.141.40* username=supervision* password=centreon* credentials

OK : Slots Busy : 10 (3.91 %), Free : 246 (96.09 %), Waiting : 9 (3.52 %), Starting : 0 (0.00 %), Reading : 0 (0.00 %), Sending : 1 (0.39 %), Keepalive : 0 (0.00 %), Dns Lookup : 0 (0.00 %), Gracefuly Finished : 0 (0.00 %), Idle Cleanup Worker : 0 (0.00 %), Closing : 0 (0.00 %), Logging : 0 (0.00 %) | 'busy'=10 ; ; ;0 ;256 'free'=246 ; ; ;0 ;256 'waiting'=9 ; ; ;0 ;256 'starting'=0 ; ; ;0 ;256 'reading'=0 ; ; ;0 ;256 'sending'=1 ; ; ;0 ;256 'keepalive'=0 ; ; ;0 ;256 'dns_lookup'=0 ; ; ;0 ;256 'gracefuly_finished'=0 ; ; ;0 ;256 'idle_cleanup_worker'=0 ; ; ;0 ;256 'closing'=0 ; ; ;0 ;256 'logging'=0 ; ; ;0 ;256

-----------------------------------------

./centreon_plugins.pl* plugin=apps::apache::serverstatus::plugin* hostname=192.168.141.40* username=supervision* password=centreon* credentials* mode=workers

OK : Busy workers : 1 Idle workers : 9 (Server Limit : 256 - 0 % Busy) | 'idle_workers'=9 ; ; ;0 ;256 'busy_workers'=1 ; ; ;0 ;256

-----------------------------------------

/plugins/centreon-plugins# ./centreon_plugins.pl* plugin=apps::apache::serverstatus::plugin* hostname=192.168.141.40* username=supervision* password=centreon* credentials* mode=requests
OK : BytesPerSec : 128.00 B AccessPerSec : 0.12 RequestPerSec : 0.02 BytesPerRequest : 29.38 KB | 'avg_RequestPerSec'=0.02 ; ; ;0 ; 'bytesPerSec'=128.00B ; ; ;0 ; 'avg_bytesPerRequest'=30082.6B ; ; ;0 ; 'avg_bytesPerSec'=527.902B ; ; ;0 ; 'accessPerSec'=0.12 ; ; ;0 ;

-----------------------------------------

./centreon_plugins.pl* plugin=apps::apache::serverstatus::plugin* hostname=192.168.141.40* username=supervision* password=centreon* credentials* mode=responsetime

OK : Response time 0.035s | 'time'=0.035s ; ; ;0 ;

---------------------------------------

./centreon_plugins.pl* plugin=apps::apache::serverstatus::plugin* hostname=192.168.141.40* username=supervision* password=centreon* credentials* mode=cpuload

OK : CPU Load : 0.00 % | 'cpuload'=0.00% ; ; ;0 ;

Configuration centreon :

Ajout d'une commande :

La commande est "check_centreon_plugin_http" :

$CENTREONPLUGINS$/centreon_plugins.pl --plugin=$_SERVICEPLUGIN$ --hostname=$HOSTADDRESS$ --mode=$_SERVICEMODE$ --username=$_SERVICEUSERNAME$ --password=$_SERVICEPASSWORD$  --warning=$_SERVICEWARNING$ --critical=$_SERVICECRITICAL$ $_SERVICEOPTION$

Ajout du modele générique :

Création : Modele-check_centreon_plugin_http

check_command ;check_centreon_plugin_http
PLUGIN ;apps::apache::serverstatus::plugin
USERNAME ;supervision
PASSWORD ;centreon ;1
OPTION ; —credentials

Azure CLI : Commandes utiles pour inventaire

Jerome ROBERT — 2019-02-14T09:50:52Z

Commande pour connaîte la liste des VM et leur IP :

az vm list-ip-addresses -o table | tr -s ' ' ' ;'

Commande pour connaîte la liste des VM ResourceGroup,Location,Zones :

az vm list -o table | tr -s ' ' ' ;'

Script pour avoir une liste des VM et IP

create-inventaire.sh
# ! /bin/bash

#fichier avec la liste des VM
az vm list -o table | tr -s ' ' ' ;' | cut -d' ;' -f1,2 | sed '1,2d' > /tmp/listVM$$
FVM=/tmp/listVM$$

#fichier avec la liste des VM/IP
az vm list-ip-addresses -o table | tr -s ' ' ' ;' | sed '1,2d' > /tmp/listIP$$
FIP=/tmp/listIP$$

for var in $(cat $FVM )
do
VM=$(echo $var | cut -d' ;' -f1)
IP=$(grep $VM $FIP | cut -d' ;' -f2)
echo $var" ;"$IP
done

cd /tmp
rm -f $FVM $FIP
exit 0

Option commande AZ :

Use az --version to display the current version.

Here are the base commands :

account : Manage Azure subscription information. acr : Manage private registries with Azure Container Registries. acs : Manage Azure Container Services. ad : Manage Azure Active Directory Graph entities needed for Role Based Access Control. advisor : Manage Azure Advisor. aks : Manage Azure Kubernetes Services. ams : Manage Azure Media Services resources. appservice : Manage App Service plans. backup : Manage Azure Backups. batch : Manage Azure Batch. batchai : Manage Batch AI resources. billing : Manage Azure Billing. bot : Manage Microsoft Bot Services. cdn : Manage Azure Content Delivery Networks (CDNs). cloud : Manage registered Azure clouds. cognitiveservices : Manage Azure Cognitive Services accounts. configure : Manage Azure CLI configuration. This command is interactive. consumption : Manage consumption of Azure resources. container : Manage Azure Container Instances. cosmosdb : Manage Azure Cosmos DB database accounts. deployment : Manage Azure Resource Manager deployments at subscription scope. disk : Manage Azure Managed Disks. dla : (PREVIEW) Manage Data Lake Analytics accounts, jobs, and catalogs. dls : (PREVIEW) Manage Data Lake Store accounts and filesystems. dms : Manage Azure Data Migration Service (DMS) instances. eventgrid : Manage Azure Event Grid topics and subscriptions. eventhubs : Manage Azure Event Hubs namespaces, eventhubs, consumergroups and georecovery configurations - Alias. extension : Manage and update CLI extensions. feature : Manage resource provider features. feedback : Send feedback to the Azure CLI Team! find : Find Azure CLI commands. functionapp : Manage function apps. group : Manage resource groups and template deployments. hdinsight : Manage HDInsight resources. identity : Managed Service Identities. image : Manage custom virtual machine images. interactive : Start interactive mode. Installs the Interactive extension if not installed already. iot : Manage Internet of Things (IoT) assets. iotcentral : Manage IoT Central assets. keyvault : Manage KeyVault keys, secrets, and certificates. kusto : Manage Azure Kusto resources. lab : Manage Azure DevTest Labs. lock : Manage Azure locks. login : Log in to Azure. logout : Log out to remove access to Azure subscriptions. managedapp : Manage template solutions provided and maintained by Independent Software Vendors (ISVs). maps : Manage Azure Maps. mariadb : Manage Azure Database for MariaDB servers. monitor : Manage the Azure Monitor Service. mysql : Manage Azure Database for MySQL servers. network : Manage Azure Network resources. openshift : (PREVIEW) Manage Azure OpenShift Services. policy : Manage resource policies. postgres : Manage Azure Database for PostgreSQL servers. provider : Manage resource providers. redis : Manage dedicated Redis caches for your Azure applications. relay : Manage Azure Relay Service namespaces, WCF relays, hybrid connections, and rules. reservations : Manage Azure Reservations. resource : Manage Azure resources. role : Manage user roles for access control with Azure Active Directory and service principals. search : Manage Azure Search services, admin keys and query keys. security : Manage your security posture with Azure Security Center. servicebus : Manage Azure Service Bus namespaces, queues, topics, subscriptions, rules and geo-disaster recovery configuration alias. sf : Manage and administer Azure Service Fabric clusters. sig : Manage shared image gallery. signalr : Manage Azure SignalR Service. snapshot : Manage point-in-time copies of managed disks, native blobs, or other snapshots. sql : Manage Azure SQL Databases and Data Warehouses. storage : Manage Azure Cloud Storage resources. tag : Manage resource tags. vm : Manage Linux or Windows virtual machines. vmss : Manage groupings of virtual machines in an Azure Virtual Machine Scale Set(VMSS). webapp : Manage web apps.

Problème avec Poller States de centreon

Jerome ROBERT — 2018-09-06T11:37:40Z

Comment faire quand nous avons "Poller States" en rouge mais que les Pollers sont OK / running ?

Généralement, cela se produit quand on a supprimé un poller.

Il va falloir ce connecter à la base de données de centreon.

Pour connaître vos Pollers actifs :

use centreon ;
SELECT name FROM nagios_server WHERE ns_activate = â ?‚ ??„ ?1â ?‚ ??„ ? ;

Pour connaître les Pollers qui posent problème :

use centreon_storage ;
SELECT last_alive AS last_update, running, name, instance_id FROM instances WHERE deleted = 0

Les Pollers qui posent problème sont en running = 0

Il faut supprimer les Pollers avec running = 0

Dans notre cas :

use centreon_storage ;
DELETE FROM instances WHERE instance_id = 5 ;
DELETE FROM instances WHERE instance_id = 6 ;
....

DELETE FROM instances WHERE instance_id = 9 ;

Résultat :

Suite à la suppression de ancien Poller TOUT est maintenant OK !!!

Les commandes SQL utilent dans ce type de probl ?ƒ¨me :

use centreon ;
SELECT name FROM nagios_server WHERE ns_activate = â ?‚ ??„ ?1â ?‚ ??„ ? ;

use centreon_storage ;
SELECT last_alive AS last_update, running, name, instance_id FROM instances WHERE deleted = 0

use centreon_storage ;
SELECT stat_value, i.instance_id, name FROM nagios_stats ns, instances i WHERE ns.stat_label = â ?‚ ??„ ?Service Check Latencyâ ?‚ ??„ ? AND ns.stat_key LIKE â ?‚ ??„ ?Averageâ ?‚ ??„ ? AND ns.instance_id = i.instance_id AND i.deleted = 0

use centreon_storage ;
DELETE FROM instances WHERE instance_id = $PourRunnig0$ ;

Gestion des droits & privilèges sous Mysql & MariaDB

Jerome ROBERT — 2018-07-23T12:03:47Z

Gestion des privilÃ¨ges sous Mysql /mariadb :

Autorise l'utilisateur "root" sur toutes les bases et tables venant de tous les "machines" (déconseillé)

grant all privileges on *.* to 'root'@'%' with grant option ;

Autorise l'utilisateur "root" sur toutes les bases et tables venant d'une IP.

grant all privileges on *.* to 'root'@'192.168.5.6' with grant option ;

Autorise l'utilisateur "root" sur toutes les bases et tables venant du serveur lui même.

grant all privileges on *.* to 'root'@'localhost' with grant option ;

Puis valider et appliquer ces opérations :

flush privileges ;

Connaître la taille des bases MariaDB / Mysql

Jerome ROBERT — 2018-06-13T10:47:32Z

Pour connaître la taille d'une base de donnée MariaDB ou Mysql :

SELECT table_schema AS "Nom de la BASE", ROUND(SUM(data_length + index_length) / 1024 / 1024, 2) AS " Taille en (MB) " FROM information_schema.TABLES GROUP BY table_schema ;

 +--------------------+-----------------+ | Nom de la BASE | Taille en (MB) | +--------------------+-----------------+ | Base1 | 98760.22 | | Base2 | 50692.66 | | information_schema | 0.17 | | mysql | 4.15 | | performance_schema | 0.00 | +--------------------+-----------------+ 5 rows in set, 1 warning (0.02 sec)

Connaître les processus qui utilisent la SWAP

Jerome ROBERT — 2018-06-07T15:06:49Z

Voici le script :

#!/bin/bash # Get current swap usage for all running processes # Usage: ./getswap.sh | sort -n -k 5 # Erik Ljungstrom 27/05/2011 mod by Poil SUM=0 OVERALL=0 for DIR in /proc/[0-9]*; do PID=${DIR##*\/} PROGNAME=$(ps -p ${PID} -o comm --no-headers) PROGPATH=$(cat /proc/${PID}/cmdline) for SWAP in $(awk '/Swap/ { print $2}' ${DIR}/smaps 2>/dev/null);do SUM=$(( SUM + SWAP )) done echo "PID=$PID - Swap used: ${SUM} - (${PROGNAME} - ${PROGPATH})" OVERALL=$(( OVERALL + SUM)) SUM=0 done echo "Overall swap used: $OVERALL"