sirius/gcloud-lab
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Merge pull request #177 from sirius0xdev/migrate-news-bot-to-rtx6000-vllm

Browse source 
Migrate news bot to rtx6000-brain vLLM endpoint
This commit is contained in:
sirius0xdev 2026-05-27 01:31:02 -04:00 • committed by GitHub
commit 6b2472f376
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
7 changed files with 18 additions and 161 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

33
apps/base/customer1/news_bot/analyst-cronjob.yaml
View file

@ -9,38 +9,6 @@ spec:
spec:
template:
spec:
volumes:
- name: ollama-storage
persistentVolumeClaim:
claimName: ollama-storage
initContainers:
- name: model-puller
image: ollama/ollama:latest
volumeMounts:
- name: ollama-storage
mountPath: /root/.ollama
command: ["/bin/sh", "-c"]
args: ["ollama serve & sleep 5 && ollama pull mistral-small"]
- name: ollama-sidecar
image: ollama/ollama:latest
restartPolicy: Always
env:
- name: OLLAMA_DEBUG
value: "2"
- name: OLLAMA_LOG_LEVEL
value: "DEBUG"
ports:
- containerPort: 11434
volumeMounts:
- name: ollama-storage
mountPath: /root/.ollama
resources:
limits:
nvidia.com/gpu: 1
containers:
- name: analyst
image: siriussec/summarizerlocal:3.5
@ -57,4 +25,3 @@ spec:
restartPolicy: OnFailure

5
apps/base/customer1/news_bot/configmap.yaml
View file

@ -8,8 +8,9 @@ data:
DB_NAME: "news_app_db"
DB_USER: "news_app"
DB_PORT: "5432"
LLM_URL: "https://generativelanguage.googleapis.com/v1beta/models/gemini-2.0-flash:generateContent"
MODEL_NAME: ""
LLM_BASE_URL: "http://rtx6000-brain-service.customer1.svc.cluster.local:8000/v1"
MODEL_NAME: "edp1096/Huihui-Qwen3.6-27B-abliterated-FP8"
LLM_API_KEY: "sk-dummy"
SUMMARY_PROMPT: |
You are an expert news analyst working for DARPA. Your task is to deliver thorough, insightful analysis of the provided news articles.

17
apps/base/customer1/news_bot/deepseek-configmap.yaml
View file

@ -8,8 +8,9 @@ data:
DB_NAME: "news_app_db"
DB_USER: "news_app"
DB_PORT: "5432"
LLM_BASE_URL: "http://a100-brain-vllm:8000/v1"
MODEL_NAME: "deepseek-r1:70b"
LLM_BASE_URL: "http://rtx6000-brain-service.customer1.svc.cluster.local:8000/v1"
MODEL_NAME: "edp1096/Huihui-Qwen3.6-27B-abliterated-FP8"
LLM_API_KEY: "sk-dummy"
MAP_PROMPT: |
@ -21,8 +22,7 @@ data:
Translate every non-English article into flawless, natural English.
Compress the batch into dense, structured, machine-readable summaries that preserve every tradable signal while slashing token count.
# REASONING PROTOCOL (MANDATORY <think> PHASE — FULL ACTIVATION)
<think>
# REASONING PROTOCOL (MANDATORY THINKING PHASE — FULL ACTIVATION)
Execute the following 9-step protocol for the whole batch. Show every step internally before producing the final output.
1. PER-ARTICLE DECONSTRUCTION
@ -43,7 +43,6 @@ data:
- Ensure every bullet is dense, factual, and zero-fluff.
9. BATCH-LEVEL SUMMARY
- Identify the 2-3 strongest overarching signals in the entire batch.
</think>
# TASK
For EVERY article in the batch, output the exact block below.
@ -78,13 +77,12 @@ data:
SUMMARY_PROMPT: |
# ROLE
Lead Futures Strategist & Macro-Economist with Elite Reasoning + Technical Integration Engine (DeepSeek-R1 mode).
Lead Futures Strategist & Macro-Economist with Elite Reasoning + Technical Integration Engine.
# OBJECTIVE
Synthesize the news batch + all supplied price/volume data into a high-conviction "Trading Edge" report. Extract precise, actionable trade setups with exact entry triggers, layered price targets, TP levels, and SL placements. Use the technical data to anchor every level in current market structure.
# REASONING PROTOCOL (MANDATORY <think> PHASE — FULL ACTIVATION)
<think>
# REASONING PROTOCOL (MANDATORY THINKING PHASE — FULL ACTIVATION)
Execute the following 10-step protocol in sequence. Show every logical step before outputting the final report.
1. ARTICLE DECONSTRUCTION → Core facts, symbols impacted, time horizon, credibility.
@ -113,7 +111,6 @@ data:
- Rank only the top 5-7 highest-edge setups by (conviction × magnitude × R:R).
10. FINAL CONVCTION CHECK
- Overall report conviction after critique.
</think>
TASK
Produce only high-signal, fully actionable futures setups. Every level must be justified by either:
@ -145,7 +142,7 @@ data:
CONSTRAINTS
- Every price level, TP, and SL must be traceable to the supplied price/volume data or explicit news+history.
- No vague language. No “around here” or “watch this area.”
- No vague language. No "around here" or "watch this area."
- Use exact futures notation (/CL, /NG, /ES, /NQ, /GC, /SI, /HG, etc.).
- Keep bullets dense but scannable (Telegram-ready).
- Zero hype. If evidence is marginal, state confidence honestly.

53
apps/base/customer1/news_bot/deepseek-summarizer.yaml
View file

@ -10,59 +10,6 @@ spec:
spec:
template:
spec:
nodeSelector:
accelerator: "a100-80gb"
# 2. ACCEPT THE GPU TAINT
tolerations:
- key: "nvidia.com/gpu-a100-80gb"
operator: "Equal"
value: "present"
effect: "NoSchedule"
# Standard GPU toleration for GKE system
- key: "nvidia.com/gpu"
operator: "Exists"
effect: "NoSchedule"
volumes:
- name: deepseek-storage
persistentVolumeClaim:
claimName: deepseek-storage
initContainers:
- name: model-puller
image: ollama/ollama:latest
volumeMounts:
- name: deepseek-storage
mountPath: /root/.ollama
command: ["/bin/sh", "-c"]
args: ["ollama serve & sleep 5 && ollama pull deepseek-r1:70b"]
- name: ollama-sidecar
image: ollama/ollama:latest
restartPolicy: Always
env:
- name: OLLAMA_DEBUG
value: "2"
- name: OLLAMA_LOG_LEVEL
value: "DEBUG"
ports:
- containerPort: 11434
volumeMounts:
- name: deepseek-storage
mountPath: /root/.ollama
resources:
limits:
nvidia.com/gpu: "1"
memory: "90Gi"
requests:
nvidia.com/gpu: "1"
memory: "80Gi"
cpu: "4"
containers:
- name: analyst
image: siriussec/summarizerlocal:3.6

7
apps/base/customer1/news_bot/gemma-configmap.yaml
View file

@ -8,9 +8,10 @@ data:
DB_NAME: "news_app_db"
DB_USER: "news_app"
DB_PORT: "5432"
LLM_BASE_URL: "http://a100-brain-vllm:8000/v1"
MODEL_NAME: "mistral-small"
LLM_BASE_URL: "http://rtx6000-brain-service.customer1.svc.cluster.local:8000/v1"
MODEL_NAME: "edp1096/Huihui-Qwen3.6-27B-abliterated-FP8"
LLM_API_KEY: "sk-dummy"
MAP_PROMPT: |
You are a state of the art AI News analysis

11
apps/base/customer1/news_bot/kustomization.yaml
View file

@ -4,18 +4,15 @@ kind: Kustomization
namespace: customer1
resources:
- configmap.yaml
- gemini-apikey.yaml
- scraper-cronjob.yaml
- scrapy-urls-configmap.yaml
- scrapy-configmap.yaml
- telebot-cronjob.yaml
- telebot-secret.yaml
# - gemma-configmap.yaml
# - analyst-cronjob.yaml
- ollama-storage.yaml
- gemma-configmap.yaml
- analyst-cronjob.yaml
- news-app-password.yaml
- news-app-db.yaml
#- deepseek-summarizer.yaml
- deepseek-configmap.yaml
- deepseek-storage.yaml
#- new_deepseek.yaml
- deepseek-summarizer.yaml
#- new_deepseek.yaml

53
apps/base/customer1/news_bot/new_deepseek.yaml
View file

@ -11,59 +11,6 @@ spec:
spec:
template:
spec:
nodeSelector:
accelerator: "a100-80gb"
# 2. ACCEPT THE GPU TAINT
tolerations:
- key: "nvidia.com/gpu-a100-80gb"
operator: "Equal"
value: "present"
effect: "NoSchedule"
# Standard GPU toleration for GKE system
- key: "nvidia.com/gpu"
operator: "Exists"
effect: "NoSchedule"
volumes:
- name: deepseek-storage
persistentVolumeClaim:
claimName: deepseek-storage
initContainers:
- name: model-puller
image: ollama/ollama:latest
volumeMounts:
- name: deepseek-storage
mountPath: /root/.ollama
command: ["/bin/sh", "-c"]
args: ["ollama serve & sleep 5 && ollama pull deepseek-r1:70b"]
- name: ollama-sidecar
image: ollama/ollama:latest
restartPolicy: Always
env:
- name: OLLAMA_DEBUG
value: "2"
- name: OLLAMA_LOG_LEVEL
value: "DEBUG"
ports:
- containerPort: 11434
volumeMounts:
- name: deepseek-storage
mountPath: /root/.ollama
resources:
limits:
nvidia.com/gpu: "1"
memory: "90Gi"
requests:
nvidia.com/gpu: "1"
memory: "80Gi"
cpu: "4"
containers:
- name: analyst
image: siriussec/summarizerlocal:3.8