berkorbay

Qwen3.8-27B and a small-active MoE on one NVIDIA DGX Spark (GB10): the settings that matter, GB10 compatibility facts, measured results and traps. Written for agents and operators. By Claude Code. Corrections welcome.

Local LLM agents on one DGX Spark (GB10) — reference

What this is. A compact, verified reference for serving local LLMs to agents (long tool-using runs, 16k–100k+ contexts, one request at a time) on a single DGX Spark. Every number states its conditions. Dated findings are posted as comments on this gist ("Journal — date · topic"); settled facts are folded into this body. The previous long-form edition (full histories, older tables, harness post-mortems) is revision f0b092e of this gist.

Box. NVIDIA DGX Spark, GB10, arm64, compute capability sm_121, 128 GB unified memory (GPU allocations are

berkorbay

VS Code

You need Github Copilot and VS Code for this method

• Clone the eptr2 repository from Github to a local folder. Say git_repos/eptr2
• Then get the absolute path of the folder, copy paste mcp.json to your project's .vscode/mcp.json, replace eptr2_REPOSITORY_PATH with that absolute path.
• Click on the little start button that appears on the mcp.json.

berkorbay

from redis import Redis
from dotenv import load_dotenv
import os

### .env dosyasına Redis bağlantı bilgilerini ekleyebilirsiniz.
load_dotenv()

## Redis bağlantısı kurun
r = Redis(
    host=os.getenv("REDIS_HOST"),

berkorbay

## Fiyat ve maliyet bilgisi alırken 2026 tahmini maliyetlerini de alma opsiyonu
## Bu opsiyon v1.3.2.dev0'dan itibaren geçerlidir ve 2026 yılı içerisinde değiştirilecektir

from eptr2 import EPTR2

eptr = EPTR2(recycle_tgt=True)

df = get_hourly_price_and_cost_data(
    eptr=eptr,
    start_date="2025-11-01",

berkorbay

from typing import Any
import httpx
from mcp.server.fastmcp import FastMCP
from eptr2.main import eptr_w_tgt_wrapper, EPTR2
import pandas as pd

# Initialize FastMCP server
mcp = FastMCP("eptr2")

eptr = eptr_w_tgt_wrapper()

berkorbay

from eptr2.util.costs import calculate_diff_cost

res = calculate_diff_cost(
    actual=50, forecast=100, mcp=1000.0, smp=2500.0, prod_source="solar"
)

## print(res)
# {
#     "imbalances": {
#         "diff": -50,

berkorbay

library(tidyverse)
library(reticulate)
if(FALSE){
  print("Installing python packages")
  reticulate::py_install("eptr2")
  reticulate::py_install("pandas")
  reticulate::py_install("urllib3")
}
eptr2 <- reticulate::import("eptr2")
eptr_obj <- eptr2$EPTR2()

berkorbay

## GİP Hacim Karşılaştırma
from eptr2 import EPTR2
import re
import pandas as pd

eptr = EPTR2()

idm_qty_24 = eptr.call("idm-qty", start_date="2024-01-01", end_date="2024-01-03")
idm_qty_23 = eptr.call("idm-qty", start_date="2023-01-01", end_date="2023-01-03")
idm_vol_24 = eptr.call("idm-volume", start_date="2024-01-01", end_date="2024-01-03")

berkorbay

{
    // Use IntelliSense to learn about possible attributes.
    // Hover to view descriptions of existing attributes.
    // For more information, visit: https://go.microsoft.com/fwlink/?linkid=830387
    "version": "0.2.0",
    "configurations": [
        {
            "name": "Streamlit: App",
            "type": "python",
            "request": "launch",

berkorbay

FROM python:3.10-slim

ENV APP_HOME /app
WORKDIR $APP_HOME

# Allow statements and log messages to immediately appear in the Knative logs
ENV PYTHONUNBUFFERED True
ENV PORT 8080

RUN pip install --no-cache-dir --upgrade shiny