Report · resources

Chat and code

Qwen3.8-27B · Mirai S trellis with Q8_0 MTP GGUF variant · Mirai S CUDA dev · with multi-token prediction

Measured results / Chat and code / Run 6c11f8fa03bf41feb5e151f4

Measured 2026-10-09 13:54 UTC · Record date 2026-10-09

Not graded. 6/6 request checks

Code writing speedcode_decode_tps · tok/s: 41.47 tok/sper person · mean · sample count not recorded
Prose writing speedchat_decode_tps · tok/s: 37.29 tok/sper person · mean · sample count not recorded
Peak memory, whole hostpeak_host_unavailable_gib · GiB: 12.61 GiBmachine total · max · sample count not recorded

Runs on this setup

Earlier runs keep their own engine versions and outcomes. Workload shapes stay separate.

History and measured sweep

History of this setup, newest first
Date	Shape	Each person · median	Machine total	First word · median	Result	Run
2026-10-09	1 people	not recorded	not recorded	not recorded	Passed	Run 858a9f76 →
2026-10-09	1 people	not recorded	not recorded	not recorded	Not graded	Run 6c11f8fa →

Answer checks

6/6 request checks

Method

Managed runtime identity checked before and after; one warmup then three 512-token requests each for code and prose, thinking off. Arithmetic mean first-content-to-DONE decode; separate warm concurrency-2 aggregate includes TTFT when present.

Limits

Speed smoke does not grade task correctness. Sampled whole-host unavailable memory is not allocator residency. Configured context is not long-context quality proof.

What ran

Exact identity

Original publisher model: Qwen3.8-27B
Publisher revision: 1d4bf0f2ff6012fd82039f2fa52739d0dd7c60c0
Variant · format · precision: Mirai S trellis with Q8_0 MTP GGUF variant · GGUF · Mirai S trellis with Q8_0 MTP
Weights SHA-256: not recorded
Engine · version · feature: Mirai S CUDA · 0.2.0-dev-vector-disabled · with multi-token prediction
Hardware: NVIDIA GeForce RTX 3060, 12 GB VRAM
Harness / workload: not recorded
Workload suite · version: not recorded
Harness SHA-256: not recorded
Fixture SHA-256: not recorded

Workload shape

People in this run: 1
Configured window (tokens): 262144
KV cache precision: q8_0
Output token ceiling: 512
Feature: with multi-token prediction

Full reproduction pack available to customers and evaluators on request

All recorded numbers

Metric definitions, scope and statistics

All aggregate numeric result fields in this run
Metric	Value	Scope	Statistic	Sample count	Basis / source field	Run
chat decode tps	37.29 tok/s	per person	mean	not recorded	chat_decode_tps	Run 6c11f8fa →
requests ok	6.00 count	machine total	single	not recorded	requests_ok	Run 6c11f8fa →
code ttft mean s	0.32 s	per person	mean	not recorded	code_ttft_mean_s	Run 6c11f8fa →
requests total	6.00 count	machine total	single	not recorded	requests_total	Run 6c11f8fa →
code decode tps	41.47 tok/s	per person	mean	not recorded	code_decode_tps	Run 6c11f8fa →
chat ttft mean s	0.30 s	per person	mean	not recorded	chat_ttft_mean_s	Run 6c11f8fa →
peak host unavailable gib	12.61 GiB	machine total	max	not recorded	peak_host_unavailable_gib	Run 6c11f8fa →

Data and citation

Download this run · OLIR JSON All numbers · CSV OLIR schema

Skilak Keep evidence, run 6c11f8fa03bf41feb5e151f4, measured 2026-10-09T13:54:43.418286+00:00.
https://skilak.ai/keep/resources/evidence/r/6c11f8fa03bf41feb5e151f4

Ready to scope it?

Start with one workload.

Tell us who uses it, what it can read, and what must stay inside. We will map the build.

Request a pilot