From a322e00659605421f4a18de1be605499a6c7743c Mon Sep 17 00:00:00 2001
From: Celes Renata <celes@frameshift.net>
Date: Fri, 3 Jul 2026 17:57:19 +0000
Subject: [PATCH] feat: vLLM serving numind/NuExtract3 on full GPU, ollama
 disabled
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

- Ollama scaled to 0 — vLLM gets entire 16GB GPU
- NuExtract3 (9B) running at 80% GPU utilization
- Both OLLAMA_BASE_URL and VLLM_BASE_URL point to vLLM service
- Fixed gremlin-1 missing default route (no internet access)
- Removed GPU resource limit (time-slicing handles allocation)
- Removed init container (direct download works with internet fix)
---
 infra/helm/stonks-oracle/values-beta.yaml  | 6 +++---
 infra/helm/stonks-oracle/values-paper.yaml | 6 +++---
 infra/helm/stonks-oracle/values.yaml       | 6 +++---
 infra/kube-vllm/deployment.yaml            | 3 ++-
 infra/kube-vllm/service.yaml               | 4 ++--
 infra/kube-vllm/vllm-metrics.yaml          | 2 +-
 6 files changed, 14 insertions(+), 13 deletions(-)

diff --git a/infra/helm/stonks-oracle/values-beta.yaml b/infra/helm/stonks-oracle/values-beta.yaml
index abf2cf6..d0a3f90 100644
--- a/infra/helm/stonks-oracle/values-beta.yaml
+++ b/infra/helm/stonks-oracle/values-beta.yaml
@@ -39,9 +39,9 @@ config:
   MINIO_SECURE: "false"
   BROKER_MODE: "paper"
   BROKER_PROVIDER: "alpaca"
-  OLLAMA_BASE_URL: "http://10.1.1.12:2701"
-  OLLAMA_MODEL: "qwen3.5:4b-fast"
-  VLLM_BASE_URL: "http://vllm-external.vllm-service.svc.cluster.local:2701"
+  OLLAMA_BASE_URL: "http://nuextract-external.vllm-service.svc.cluster.local:2701"
+  OLLAMA_MODEL: "numind/NuExtract3"
+  VLLM_BASE_URL: "http://nuextract-external.vllm-service.svc.cluster.local:2701"
   VLLM_MODEL: "numind/NuExtract3"
   VLLM_TIMEOUT: "120"
   VLLM_MAX_RETRIES: "2"
diff --git a/infra/helm/stonks-oracle/values-paper.yaml b/infra/helm/stonks-oracle/values-paper.yaml
index 43c7d6b..cc78d97 100644
--- a/infra/helm/stonks-oracle/values-paper.yaml
+++ b/infra/helm/stonks-oracle/values-paper.yaml
@@ -16,9 +16,9 @@ config:
   REDIS_DB: "2"
   DEPLOY_STAGE: "paper"
   POSTGRES_USER: "stonks_paper"
-  OLLAMA_BASE_URL: "http://10.1.1.12:2701"
-  OLLAMA_MODEL: "qwen3.5:4b-fast"
-  VLLM_BASE_URL: "http://vllm-external.vllm-service.svc.cluster.local:2701"
+  OLLAMA_BASE_URL: "http://nuextract-external.vllm-service.svc.cluster.local:2701"
+  OLLAMA_MODEL: "numind/NuExtract3"
+  VLLM_BASE_URL: "http://nuextract-external.vllm-service.svc.cluster.local:2701"
   VLLM_MODEL: "numind/NuExtract3"
   MARKET_DATA_BASE_URL: "https://api.polygon.io"
 
diff --git a/infra/helm/stonks-oracle/values.yaml b/infra/helm/stonks-oracle/values.yaml
index 66d959d..0c5b30a 100644
--- a/infra/helm/stonks-oracle/values.yaml
+++ b/infra/helm/stonks-oracle/values.yaml
@@ -185,14 +185,14 @@ config:
   REDIS_DB: "0"
   MINIO_ENDPOINT: "minio.minio-service.svc.cluster.local:80"
   MINIO_SECURE: "false"
-  OLLAMA_BASE_URL: "http://10.1.1.12:2701"
-  OLLAMA_MODEL: "qwen3.5:4b-fast"
+  OLLAMA_BASE_URL: "http://nuextract-external.vllm-service.svc.cluster.local:2701"
+  OLLAMA_MODEL: "numind/NuExtract3"
   OLLAMA_TIMEOUT: "240"
   OLLAMA_MAX_RETRIES: "2"
   OLLAMA_RETRY_BASE_DELAY: "1.0"
   OLLAMA_RETRY_MAX_DELAY: "10.0"
   OLLAMA_RETRY_BACKOFF_MULTIPLIER: "2.0"
-  VLLM_BASE_URL: "http://vllm-external.vllm-service.svc.cluster.local:2701"
+  VLLM_BASE_URL: "http://nuextract-external.vllm-service.svc.cluster.local:2701"
   VLLM_MODEL: "numind/NuExtract3"
   VLLM_TIMEOUT: "120"
   VLLM_MAX_RETRIES: "2"
diff --git a/infra/kube-vllm/deployment.yaml b/infra/kube-vllm/deployment.yaml
index 53060c3..9014954 100644
--- a/infra/kube-vllm/deployment.yaml
+++ b/infra/kube-vllm/deployment.yaml
@@ -16,6 +16,7 @@ spec:
         app: vllm
     spec:
       runtimeClassName: nvidia
+      enableServiceLinks: false
       nodeSelector:
         kubernetes.io/hostname: gremlin-1
       containers:
@@ -32,7 +33,7 @@ spec:
             - "--port"
             - "8000"
             - "--gpu-memory-utilization"
-            - "0.45"
+            - "0.80"
             - "--max-model-len"
             - "8192"
             - "--max-num-seqs"
diff --git a/infra/kube-vllm/service.yaml b/infra/kube-vllm/service.yaml
index 446c319..4bdc350 100644
--- a/infra/kube-vllm/service.yaml
+++ b/infra/kube-vllm/service.yaml
@@ -1,7 +1,7 @@
 apiVersion: v1
 kind: Service
 metadata:
-  name: vllm
+  name: nuextract
   namespace: vllm-service
 spec:
   ports:
@@ -16,7 +16,7 @@ spec:
 apiVersion: v1
 kind: Service
 metadata:
-  name: vllm-external
+  name: nuextract-external
   namespace: vllm-service
 spec:
   ports:
diff --git a/infra/kube-vllm/vllm-metrics.yaml b/infra/kube-vllm/vllm-metrics.yaml
index f894d9f..6c61eef 100644
--- a/infra/kube-vllm/vllm-metrics.yaml
+++ b/infra/kube-vllm/vllm-metrics.yaml
@@ -37,7 +37,7 @@ metadata:
 data:
   default.conf: |
     upstream vllm_backend {
-        server vllm.vllm-service.svc.cluster.local:8000;
+        server nuextract.vllm-service.svc.cluster.local:8000;
     }
 
     server {