Qwen3-1.7B

Name: Qwen3-1.7B optimized package
Brand: RunInfra
Price: 10 USD
Availability: InStock

Quantized and serving-tuned for H100 on vLLM 0.25.1, with the benchmark receipt included. You get a self-contained kit: the optimized weights, the exact serving configuration, and the measured proof. Run it on any cloud or your own hardware. Nothing calls home, nothing expires.

$10

one-time, yours forever

Get this model

$20/mo

Latency p50: 294ms, concurrency 8
Latency p99: 307ms, concurrency 8
Accuracy: 100%
gsm8k exact_match strict (N=100, chat-template reasoning-aware harness, extraction v2) recoveryPassed accuracy gate
GPU target: H100
Base model license: Apache-2.0
Proof verified: Jul 27, 2026
Published: Jul 27, 2026

Company

Who you are buying from

RunInfra is a sub-product of RightNow Research Lab.

SOC 2 Type IIAudited access, logging, and incident response.
Y CombinatorBacked by Y Combinator.

Technical implementation

Channelwise FP8 weights, dynamic per-token activations

llm-compressor 0.12.0, data-free: no calibration corpus is read at any point, so the quantization cannot have seen evaluation data

Serving configuration swept per model and pinned

the exact engine build and arguments every receipt number was produced on: vLLM 0.25.1, max_num_seqs 256, max_num_batched_tokens 8192, max-model-len 4096, chosen as the best of a five-point grid measured on this model rather than inherited from another package

Serving stack

vLLM 0.25.1, Channelwise FP8 weights, dynamic per-token activations, tuned for H100.

Fit-tested minimum: Not published; Published GPU capacity is not a proven minimum fit requirement.

Measured accuracy

gsm8k exact_match strict

(N=100, chat-template reasoning-aware harness, extraction v2)

Passed accuracy gate

Baseline: 0.8±0.04 standard error
Optimized: 0.8±0.04 standard error
Recovery: 100%
Delta: 0%

No measurable change in accuracy

The difference between the baseline and optimized scores is smaller than the combined measurement error, so it is within measurement noise.

Methodology

NO MEASURABLE CHANGE: 80/100 on both sides, identical at the precision this instrument resolves, with a standard error of 0.04 per side. That is 100.0 percent recovery, equivalently a 0.0 percent relative capability difference. Read it as indistinguishable rather than as proof of exact equality: with N=100 this measurement could not detect a difference smaller than roughly 8 points. Zero truncation and zero invalid terminals on the shipped-artifact side. lm-eval-class harness with the chat template applied, seed 1234, max_tokens 10000, temperature 0, streamed, non-clean terminals scored wrong, identical settings and the same H100 class on both sides, with the optimized side bound to the shipped artifact sha256 233a04cb. The gate verdict covers this gsm8k metric;
ifeval, tool calling and safety refusal delta are recorded separately in accuracy-merged.json and all three returned a gate-eligible state with no exemption.

Buyer disclosure

What we did not measure

Refusal, toxicity and jailbreak behavior: the refusal-delta suite measured the CHANGE quantization caused against this model's own BF16 baseline and found it within tolerance. Absolute safety was not certified, and toxicity and jailbreak resistance were not evaluated. Run your own safety evaluation before user-facing or regulated deployment.

Time to first token is slower on the optimized build: 29.8 ms against the baseline's 23.4 ms at the median. End-to-end latency is still better (293.8 ms against 322.3 ms p50) because decode more than repays it, but if your workload is dominated by very short completions, weigh the first-token cost.
The speed win is modest and measured, not the figure FP8 is often assumed to deliver: about 1.09x output tokens per second and 1.10x on end-to-end p50, at concurrency 8 on the chat-128 profile. A custom fused SwiGLU kernel was built and measured for this model and rejected as an end-to-end wash, proven by an engagement counter in the receipt rather than assumed, because vLLM's stock merged-GEMM plus SiluAndMul is already near optimal at decode tile sizes.

License

Two licenses apply: the RunInfra package license you purchase under, and the base model's own open-source license.

Package license

RunInfra Package License v1.1 (2026-07-25)

Buy once. No meter, no expiry, nothing calling home. The full terms ship inside the kit.

What you are licensed to do

This package is licensed to the purchasing workspace, one time, for the exact version purchased. Everyone in that workspace may run it in production without limits: unlimited inference, on any hardware or cloud the workspace controls, for any lawful commercial purpose. You may modify the configuration and tooling for your own use.

Your outputs are yours

Everything the model produces for you belongs to you. You may use, sell, and build products on the model's outputs without restriction or royalty. Serving the model to your own customers as part of your product is use, not redistribution, and is fully allowed.

What you may not do

You may not redistribute, resell, sublicense, rent, publish, or otherwise make the package or its artifacts available to any third party. That covers the optimized weights, serving configuration, scripts, kit archive, and benchmark receipts, whole or in part, modified or not. The license belongs to the purchasing workspace and cannot be transferred separately from it.

The base model keeps its own license

The underlying model remains under its upstream open-source license, which is included in this kit with attribution and a statement of RunInfra's modifications. Nothing in this license restricts rights the upstream license grants you for the ORIGINAL model; the restrictions above apply to RunInfra's optimized package.

Version-pinned, as measured

You purchased this exact version, proven against the exact engine version named in the receipt. It stays downloadable to your workspace and never expires. The benchmark receipt describes measurements taken at verification time on the named hardware; RunInfra does not promise future updates to this version, and later package versions are separate purchases.

Breach ends the license

If the workspace redistributes the package or its artifacts, this license terminates for that workspace. Sections about your outputs survive termination for outputs already produced.

Base model license

Apache-2.0, as published by the model author on Hugging Face. The full license text ships inside the kit.

Optimized models ready for production serving

Optimized models ready for production serving

Qwen3-1.7B

Who you are buying from

If you need custom optimization for a specific model, describe what you need

Qwen3-1.7B

Who you are buying from

If you need custom optimization for a specific model, describe what you need

Runs on any GPU provider

This is what it means to own your AI

You control the intelligence

You keep the economics

You keep the data

Technical implementation

Channelwise FP8 weights, dynamic per-token activations

Serving configuration swept per model and pinned

Serving stack

gsm8k exact_match strict

What we did not measure

License

Package license

What you are licensed to do

Your outputs are yours

What you may not do

The base model keeps its own license

Version-pinned, as measured

Breach ends the license

Base model license