Hacktoberfest 2026: the issues maintainers tagged for October, open and beginner-friendly. Browse Hacktoberfest issues

[Benchmark] RTX 4090 24GB and 256GB system RAM. unsloth/Qwen3.8-Flash-Next IQ4_XS

Open Beginner friendly
  • Is something wrong?
#3,197 3 comments 0 reactions 0 assignees View on GitHub

Maintainers usually reply within 3 days

Nobody has claimed this yet.

Assessment

Difficulty
1/5
Estimated time
Under an hour
Newbie friendliness
64/100
Issue type
Documentation
Clarity
Clearly specified
Activity status
Active
Tech stack
cpp, linux
Domain
machine-learning, performance

Research direction

This issue records a speed benchmark using tools/<key>_bench.py and links result.json; start by reviewing the benchmark data and the linked results. Check that the reported hardware, configuration, method, and speed measurements are represented accurately, and that the stated limitations make clear that correctness was not measured.

Written by the indexing model from the issue text.

Description

Community benchmark on NVIDIA GeForce RTX 4090

Measured on 2026-10-19 with tools/<key>_bench.py (suite 1, 3 runs per test, medians with ranges). Speeds are the engine's own timings; first-token times are measured by the client over HTTP. Speed only: no correctness check was run.

Hardware and software

  • GPU and VRAM: NVIDIA GeForce RTX 4090, 24 GiB VRAM
  • CPU: AMD EPYC 7443P 24-Core Processor (24 cores, 24 threads); installed RAM: 251.5 GiB; storage: not measured
  • OS: Linux 6.12.113+deb13-amd64; GPU driver: 615.71.09; CUDA: not read
  • : project 0.1.45, commit not a git checkout; engine 0.1.45
  • Background workloads and power limits: not recorded

Model and configuration

  • Model and quantization: qwen3.8-flash-next-unsloth-ud-iq4_xs
  • Context: 100000 tokens; parallel slots: 1
  • Settings: the install's own configuration, unchanged; thinking off for the suite; temperature 0
<key> --pack unsloth-ud-iq4_xs --native Qwen3.8-Flash-Next-UD-IQ4_XS-00001-of-00003.gguf --expert-profile expert-profile.bin --expert-cache auto --prefill auto --spec 4 --spec-min-p 0.5 --mtp rt --max-context 100000 --kv int8 --kv-resident 32768 --vision --vram-reserve-mib 700

Environment of the config: <key>_HC_Q8=1, <key>_ROUTE_TAIL_SKIP=0

Method

Run by python tools/<key>_bench.py against the HTTP API of a server (already running); model loading is not timed. One warm-up chat and a 1K prompt are discarded first. Every prompt starts with a unique tag, so the conversation cache cannot help the fresh-prompt tests.

  • short chat: the three questions of tools/ab_engine.py, up to 100 new tokens, 3 runs
  • 4K / 16K prompts: generated word lists sized by a calibration request, 24 new tokens, 3 fresh prompts each
  • agent session: system prompt and four tools, then 5 turns that each add about 1500 words of a project file and continue from the model's own earlier answers (the way agent clients resend a conversation), 64 new tokens per turn, 3 sessions; the cache numbers come from the answer's usage (cached_tokens) and timings (prompt_n)
  • concurrency (when run): N requests sent at the same moment, total tokens over the wall time of the burst, latency over all requests

Results

Configuration Actual prompt tokens Reused tokens Generated tokens Runs Prompt tok/s median and range Decode tok/s median and range TTFT seconds median and range
short chat 46 (43-51) 0 100 (100-100) 9 not measured 103.8 (96.2-108.3) 0.30 (0.25-0.33)
4K prompt, fresh 4019 (4019-4019) 0 (0-0) 24 cap 3 1943.9 (1937.7-1945.6) not measured 2.09 (2.09-2.09)
16K prompt, fresh 16058 (16058-16058) 0 (0-0) 24 cap 3 3493.6 (3490.3-3501.1) not measured 4.63 (4.62-4.63)
agent session, follow-up turns 14590 (14556-14612) at the last turn see below up to 64 3 see below 117.9 (91.9-138.4) 2.16 (2.03-2.35)

Ranges are the lowest and highest of the runs; Runs counts requests that returned a token.

Agent session (conversation cache)

Cache hit over the follow-up turns: 68.6 % (68.5-68.7); earlier-turn tokens read again per session: 74 (73-145) (earlier turns that the cache did not hold; the 13718 (13717-13721) new tokens of the files are read once); first turn (nothing cached) first token: 0.90 s (0.87-0.91).

Turn (run 1) Prompt tokens Reused Read Prompt tok/s First token (s) Decode tok/s
0 609 0 609 696.0 0.895 80.6
1 3757 634 3123 1467.5 2.144 102.9
2 5290 3750 1540 760.9 2.049 98.3
3 8723 5353 3370 1566.8 2.169 103.0
4 11649 8716 2933 1259.1 2.348 132.7
5 14612 11712 2900 1355.4 2.163 107.1

Per-request data: results.json. Engine log lines from the start: engine-log-excerpt.txt (when this run started the server).

Correctness and limitations

Not measured: this is a speed suite. The conversation uses the model's own replies, so token counts differ a little between runs. Machine details come from the server, nvidia-smi/rocm-smi and the OS; background load and power limits are not recorded. Total time of the suite: 78 s.

result.json

Dominant language
C++
Stars
11.6k
Forks
1k
No merged PRs in 30d

Getting set up

This project ships no dev container, Dockerfile or contributing guide, so setting up is up to you: start from its README, and see our first-contribution guide for the general steps.

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

More from Niko1221/Strata

All issues in Niko1221/Strata

Similar issues

More C++ issues

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.