Mia's one-click kit: Qwen3.8-27B, EXL3 3.0 bpw + MTP

What it builds as an agent

Task: Airship dock (3D scene, one HTML file). One attempt, no help. Counted: the final file counts.

The browser check

• page loads without page error1 / 1
• window.ready === true within 30 s1 / 1
• no console errors (whole session)1 / 1
• frame is not blank1 / 1
• frame rate in software GL (rAF over 5 s)0 / 1
• required objects (voxel mass, >=3 steam emitters, >=2 airships, label)1 / 2
• motion: pixel diff 0.5, >=3 rotating nodes 0.5, large group translating 12 / 2
• OrbitControls: drag and wheel change the picture beyond the animation baseline0 / 1

Measured 2026-10-20. Raw files: check · run info · command · final choice · contract · sampling · screenshot · clip.mp4.

The run's own headless check browser, which renders the page the model built, used up to 1497 % CPU. The load monitor counted it as another process. Hard stop: the agent was still working when the time limit of 120 minutes was reached (7,202 s, 106 requests, 89,206 tokens written), so the harness was stopped. The last version of the page that loaded without errors counts: the final file (99,761 bytes), which also loaded cleanly in the check. Browser check 7.0 out of 10: the page loads, runs without console errors, is not blank, shows the label and animates. The three lost points: frame rate (3.7 frames per second in software rendering without a GPU), required objects (the check found the voxel mass and the label but counted no steam emitters and no airships; the count is a heuristic and can miss them) and mouse control (drag and wheel changed the picture less than the check requires). Earlier attempts on this setup scored 0 of 10 because of faults in our test chain (thinking temperature 0.6 instead of the model card's 1.0; the harness gave up after 300 s of silence from the server) and are not shown. The test gate between harness and server sent an empty keepalive every 30 s while the server was silent; the harness ignores it, model, prompt, settings and time limits are unchanged. Thinking sampling is the model card's (temperature 1.0 / top_p 0.95 / top_k 20), not the kit's default 0.6. The answer limit is the model card's 262,144 tokens (the kit's own harness default is 65,536); with a context of 117,760 the real ceiling per answer is the room left in the context, and the gate and the kit server clamp each request to it.

run_showcase.sh --harness dsh-a2 --task K1 --upstream http://127.0.0.1:38605/v1 --model qwen3.8-27b-exl3-3.0bpw --sampling-file $RUN/sampling-show.json --out $RUN/raw/plan-2010-show-r2/show --limit-s 7200 --inject-after-s 5400 --poll-s 20 --ctx 117760

How fast it runs

Method: code+prose; 2 run(s) each; 512 tokens; 1 warm-up(s); median. Measured 2026-10-20. Raw files: speed result · speed contract · log · long answer.

python3 speed2.py --base-url http://127.0.0.1:41285/v1 --label run-speed-r5 --prompts $BENCH/repos/rigmark-v2/prompts-v2.json --output $RUN/raw/plan-2010-speed-r5/work/speed2 --contract $RUN/raw/plan-2010-speed-r5/work/speed2/speed-contract.json --parts decode,prefill,conc,l1 --prefill-depths 2048,8192,32768 --prompt-ids code,prose --runs 2 --max-tokens 512 --warmups 1 --prefill-runs 3 --tokenizer-json $HOME/tp2-bench/models/Qwen3.8-27B-EXL3-4.0bpw/tokenizer.json --tokenizer-python $BENCH/venv/bin/python --sampling {"temperature": 0.7, "top_p": 0.8, "top_k": 20, "min_p": 0.0, "presence_penalty": 1.5} --sampling-source thinking = Mia kit server defaults: MiaAI-Lab/Qwen3.8-27B-16gb-NVIDIA-GPUs-one-click-install@bbb11db:README.md:497 ("Defaults: temperature 0.6, top-p 0.95, top-k 20, thinking on") and tools/serve_openai.py:961-963 (request values win; min_p and presence_penalty are not read by the kit server, 0.0 = Qwen model card). instruct (speed2 etc., thinking off) = Qwen model card as in the card box-5-qwen38-27b-exl3-3.0bpw-kit-5060ti: https://huggingface.co/Qwen/Qwen3.8-27B README.md line 253. --model qwen3.8-27b-exl3-3.0bpw --no-cache-prompt --server-log $RUN/logs/server-A.log --engine-version-file $RUN/logs/engine-version.txt

How well it answers

Thinking off, except the last test. Seed 42. Sampling from the model card. A cut-off answer counts as wrong. Measured 2026-10-20.

bash tools/quality2/measure_all.sh tools/quality2/cards/qwen38-27b-exl3-3.0bpw-kit-dsh-5060ti.json --parts mmlupro --job plan-2010

bash tools/quality2/measure_all.sh tools/quality2/cards/qwen38-27b-exl3-3.0bpw-kit-dsh-5060ti.json --parts gpqa --job plan-2010

bash tools/quality2/measure_all.sh tools/quality2/cards/qwen38-27b-exl3-3.0bpw-kit-dsh-5060ti.json --parts ifeval --job plan-2010

bash tools/quality2/measure_all.sh tools/quality2/cards/qwen38-27b-exl3-3.0bpw-kit-dsh-5060ti.json --parts lcb --job plan-2010

bash tools/quality2/measure_all.sh tools/quality2/cards/qwen38-27b-exl3-3.0bpw-kit-dsh-5060ti.json --parts ruler --job plan-2010

bash tools/quality2/measure_all.sh tools/quality2/cards/qwen38-27b-exl3-3.0bpw-kit-dsh-5060ti.json --parts ruler --job plan-2010

bash tools/quality2/measure_all.sh tools/quality2/cards/qwen38-27b-exl3-3.0bpw-kit-dsh-5060ti.json --parts think --job plan-2010

How much context fits

Memory test. For each context size (4K, 8K, 16K ... up to the largest size tested for the setup) the server is started again with that context, one prompt that fills about 95 % of the context is sent, and the model must answer. The largest size that loaded and answered counts. Decode speed is measured on that answer at the largest size. Peak memory is the GPU memory in use (nvidia-smi); the GPU was empty before each start. Measured 2026-10-20. Raw files: result of every step · log · contract.

Note from the run: Highest TESTED step is 117760 tokens = the context of the kit's 3.0 bpw profile (steps 4096 to 117760 all loaded and answered, first_failure none); a larger context was not tested, so 117760 is the profile, not a proven hardware limit. Prompt filled to 96.19 % (113268 tokens). Peak 15.19 GiB (nvidia-smi, GPU 1) at the largest step; decode 38.78 tok/s at that fill (at 4K-32K fill: 45.1-51.2 tok/s). The list of sizes ends at 117,760 tokens, which is the context of the kit's 3.0 bpw profile (all steps from 4,096 to 117,760 loaded and answered), so this is the profile's size, not a proven hardware limit; a larger context was not tested. Decode speed at this fill is 38.8 tok/s (45 to 51 tok/s at 4K to 32K).

bash tools/quality2/measure_all.sh tools/quality2/cards/qwen38-27b-exl3-3.0bpw-kit-dsh-5060ti.json --parts fit --job plan-2010

Also made this possible

• ModelQwenHugging Face @Alibaba_Qwen GitHub Website
• Quant (EXL3)turboderp@turboderp_ Hugging Face GitHub

Run it yourself

What you need and how to start it, step by step. The results below come from exactly this setup.

What you need

Hardware

1x RTX 5060 Ti 16 GB

System

Linux (we used the Linux start script)

Model

turboderp/Qwen3.8-27B-exl3 3.00bpw

Quant (EXL3)

huggingface.co/turboderp/Qwen3.8-27B-exl3

Engine

ExLlamaV3 1.6.0 — the program that loads the model and answers requestsBuild and settings we used: ExLlamaV3 1.6.0 (Mia one-click kit bbb11db), MTP

Recipe (one-click kit)

github.com/MiaAI-Lab/Qwen3.8-27B-16gb-NVIDIA-GPUs-one-click-install by Mia (MiaAI-Lab)

Steps

1. Get the recipe and the model files from the list above. We ran the engine, ExLlamaV3 1.6.0, with the recipe (Mia one-click kit bbb11db).

2. Put these settings into the recipe's .env file. One setting per line.

   Copy

   MODEL_DIR=Qwen3.8-27B-EXL3-3.0bpw
   CONTEXT_SIZE=117760
   CACHE_QUANT=4
   GPU_MEM_GB=14.7
   DRAFT=mtp
   PROFILE=3.0bpw-118k

3. Start the server with this command. It loads the model into memory.

   Copy

   CUDA_DEVICE_ORDER=PCI_BUS_ID \
     CUDA_VISIBLE_DEVICES=1 bash linux/start.sh \
     --no-harness \
     --no-pick

   What the parts mean

   • CUDA_DEVICE_ORDER=PCI_BUS_IDcounts the GPUs in slot order, the same order nvidia-smi shows
   • CUDA_VISIBLE_DEVICES=1which NVIDIA GPU the program may use: the number of the GPU in our machine (with one GPU it is 0)
   • MODEL_DIR=Qwen3.8-27B-EXL3-3.0bpwthe folder with the model files
   • CONTEXT_SIZE=117760the context window in tokens: how much text the model can see at once
   • DRAFT=mtpMTP (multi-token prediction): the model guesses several tokens ahead and checks them, which makes answers faster
   • CACHE_QUANT=4keeps the context memory (KV cache) in 4 bits, which uses less GPU memory
   • GPU_MEM_GB=14.7how much GPU memory, in GB, the kit plans with
   • PROFILE=3.0bpw-118kthe name of the kit profile these settings belong to; the name gives the bits per weight and the context size

ExLlamaV3 1.6.0 (Mia one-click kit bbb11db), MTP; CUDA_DEVICE_ORDER=PCI_BUS_ID CUDA_VISIBLE_DEVICES=1 bash linux/start.sh --no-harness --no-pick (.env: MODEL_DIR=Qwen3.8-27B-EXL3-3.0bpw CONTEXT_SIZE=117760 CACHE_QUANT=4 GPU_MEM_GB=14.7 DRAFT=mtp PROFILE=3.0bpw-118k)

Mia's own benchmark, run unchanged

This is Mia's own benchmark tool (commit 21a3511), run unchanged. It has its own section here and its own column on the class page, and it is never mixed with my numbers. A cut-off answer counts as wrong here, so a value can differ from sparkDash's own screen.

Seed 1234 applies to the two quality tests; decode, prefill and tool eval send no seed. The benchmark client ran on a separate machine that is never part of the measured group. Measured 2026-10-20. Version 2.50.0, commit 21a3511.

QA. Quality QA, 150 items, seed 1234, temperature 0, thinking off, 4 requests at once (Mia's default). An answer cut off at the token limit counts as wrong. ± is the 95 % margin (Wilson).

MMLU. Quality MMLU, 285 items (5 for each of 57 subjects, 4 answers), seed 1234, temperature 0, thinking off. An answer cut off at the token limit counts as wrong. ± is the 95 % margin (Wilson).

Decode. Decode bench, 1 request at once, 400 tokens, prompt type structured, temperature 0, thinking off. No seed is sent.

Prefill. Prefill bench at 4K, 8K, 16K and 32K tokens. sparkDash skips a size that does not fit into the context. No seed is sent.

Tool Eval Quick. Tool Eval Quick: tool-eval-bench (SeraphimSerapis, commit c61319b), 15 core scenarios (--short), --no-think, --timeout 120. Score 0 to 100. No seed is sent.

Note from the run: sparkDash 2.50.0 (@21a3511) on a client machine (never in the measured group), Quality: seed 1234, temperature 0, thinking off, a truncated answer counts as wrong; decode, prefill and tool eval send no seed; own row, never mixed with quality-v2.

python3 tools/quality2/sparkdash_client.py tools/quality2/cards/qwen38-27b-exl3-3.0bpw-kit-dsh-5060ti.json <box> --client <client machine> --job plan-2010 --out $RUN

tool-eval-bench --base-url http://127.0.0.1:<port> --model qwen3.8-27b-exl3-3.0bpw --short --no-think --timeout 120 --json-file $BENCH/sparkdash-client/runs/work/tooleval/result.json --no-live