Qwen3 × LM Studio: your first local chat
From official4B/8B GGUF to file checks,import preview and first chat. Original guide; not runtime-tested by this site.
Model to check
Models generate and reason. Agents use tools to work on tasks. A harness manages an agent’s environment. MCP connects tools and data. Skills are reusable task guidance.
Research notes
Goal
Complete a first local text chat using a verified Qwen3 file.
Who this is for
For first-time local-model users. Most steps use LM Studio’s interface; the code example is Windows PowerShell only.
Before you start
Prerequisites
Need compatible LM Studio,sufficient disk/RAM and verified GGUF. Record app/runtime versions; no fixed minimum or speed guarantee.
Model / service
Only official Qwen3-4B/8B GGUF in Q4_K_M or Q8_0; see model details for pinned revision,hash and capacity.
Agent / runtime
Start inside LM Studio without an agent,MCP or HTTP server. Later integrations need separate permission and data-flow checks.
Estimated time
Learner time varies by environment; no measured duration is claimed.
Implementation steps
Check capacity and runtime first
Stage practice: Open hardware checks from the 4B or 8B detail below. Keep context 4096 and your chosen quantization; confirm LM Studio system/runtime support separately.
Pin the official model and file
Stage practice: Use Qwen’s repository and distinguish 4B from 8B. Select Q4_K_M or Q8_0; pinned links, bytes, SHA256 and original review dates are on model pages. File size is not runtime memory.
Set the storage location first
Stage practice: Install from the official channel and open LM Studio. Set a model folder with space in My Models before downloading. Example: G:\code\models; adapt it and allow room for both original and copied file.
Sources for this stage: LM Studio · Download an LLMVerify the file and preview import
Stage practice: The Windows PowerShell preview defaults to4B/Q4_K_M. Change the path and verify SHA; another model/quant requires its own hash. Check local lms help for --dry-run/--copy and inspect the target. This snippet does not copy.
Copy on import after review
Stage practice: After checking the target, separately run the printed lms import command with --copy and read its prompts. Without a copy option it moves the file. On macOS/Linux adapt paths using official docs instead of the Windows snippet.
Load and run a short-text check
Stage practice: In Chat’s loader select the imported file, start at4096 context and wait. Ask with synthetic text: “Reply only READY /no_think”. Record output,time and memory; a format miss alone does not prove corruption. No API server is needed first.
Record results and failure boundaries
Stage practice: Record app/runtime versions,file/SHA,context,GPU offload and errors. For memory failures unload other models,reduce context or recheck4B. For unknown formats check the runtime. Prepare all assets before offline retests; check remote tools/MCP separately.
Sources for this stage: LM Studio · offline operation
Commands and examples
Not tested by AQI AI
# Windows PowerShell: verify a prepared Qwen3-4B Q4_K_M file, then preview only.
$aqiFile = 'G:\code\models\Qwen3-4B-Q4_K_M.gguf'
$aqiExpected = '7485fe6f11af29433bc51cab58009521f205840f5b4ae3a32fa7f92e8534fdf5'
if (-not (Test-Path -LiteralPath $aqiFile -PathType Leaf)) { throw 'Set the path to your verified local GGUF file.' }
if ((Get-FileHash -LiteralPath $aqiFile -Algorithm SHA256).Hash.ToLowerInvariant() -ne $aqiExpected) { throw 'SHA256 mismatch: stop and check the model source.' }
$aqiHelp = (& lms import --help | Out-String)
if ($LASTEXITCODE -ne 0 -or $aqiHelp -notmatch '--dry-run' -or $aqiHelp -notmatch '--copy') { throw 'Check your local lms version and official import documentation.' }
& lms import $aqiFile --dry-run
if ($LASTEXITCODE -ne 0) { throw 'Import preview failed; keep the error and review it.' }
# Review the preview destination before running the command printed below separately.
Write-Output ('lms import "' + $aqiFile + '" --copy')Validation checks
Verify yourself:matching SHA,correct preview destination,visible in My Models,completed load and text output. Import success does not prove every feature/device; retain failures.
Recorded test results
Official documentation reviewed; this site has not run this combination or tested the device.
Common problems
lms missing:open the app and check CLI docs. SHA mismatch:stop. Model absent:check current model directory,complete file and layout. Missing runtime:prepare the matching official runtime before retrying.
Cost
Not verified
Price not verified; check software, model, API, hardware and hosting costs separately.
Limits & boundaries
Permissions & risks
Check agents, MCP, web search and remote APIs separately. These steps are not an isolation test of your device by this site.
Alternatives and next steps
Open-source alternatives may still require paid hardware or model calls.
Prefer public software on existing hardware; API, hardware, electricity and hosting are not guaranteed free.
Original sources
- LM Studio · Download an LLM ↗
lmstudio.ai · · Official documentation
- LM Studio · Import Models ↗
lmstudio.ai · · Official documentation
- LM Studio · lms import ↗
lmstudio.ai · · Official documentation
- LM Studio · Get started ↗
lmstudio.ai · · Official documentation
- Qwen · LM Studio ↗
qwen.readthedocs.io · · Official documentation
- LM Studio · lms CLI ↗
lmstudio.ai · · Official documentation
- Qwen/Qwen3-4B-GGUF ↗
huggingface.co · · Official documentation
- Qwen/Qwen3-8B-GGUF ↗
huggingface.co · · Official documentation
- LM Studio · offline operation ↗
lmstudio.ai · · Official documentation