Starlings.Models.Qwen35
0.5.0
dotnet add package Starlings.Models.Qwen35 --version 0.5.0
NuGet\Install-Package Starlings.Models.Qwen35 -Version 0.5.0
<PackageReference Include="Starlings.Models.Qwen35" Version="0.5.0" />
<PackageVersion Include="Starlings.Models.Qwen35" Version="0.5.0" />
<PackageReference Include="Starlings.Models.Qwen35" />
paket add Starlings.Models.Qwen35 --version 0.5.0
#r "nuget: Starlings.Models.Qwen35, 0.5.0"
#:package Starlings.Models.Qwen35@0.5.0
#addin nuget:?package=Starlings.Models.Qwen35&version=0.5.0
#tool nuget:?package=Starlings.Models.Qwen35&version=0.5.0
Starlings.Models.Qwen35
The model definition of Qwen/Qwen3.5-0.8B at revision 2fc06364715b967f1860aea9cf38778875588b17, without a runtime: what any backend needs to prompt the model and read its output. No ONNX Runtime dependency (only Starlings.Generation.Abstractions, Starlings.Acquisition and the owned tokenizer Starlings.Tokenizers). To run the model in-process on ONNX Runtime GenAI, reference Starlings.Models.Qwen35.OnnxGenAI, which brings this package and holds the Stable contract of the three registry profiles.
Qwen35ChatTemplate: the pinned chat template rendered in C# for text and tool-calling conversations (Qwen35Message,Qwen35Tool,Qwen35ToolCall,enable_thinking);Qwen35ChatTemplate.Instanceis the same renderer behindIChatTemplateover the backend-neutralChatTurnandToolDefinition.Qwen35ChatTemplate.EscapeSpecialTokens(and theescapeSpecialTokensargument ofRender, opt-in) rewrites every<|name|>in message content to<¦name¦>, so content cannot rewrite the conversation; by default content is rendered verbatim, as upstreamapply_chat_templatedoes.Qwen35ToolCallParser: the template's tool-call format, whole or streamed;Qwen35ToolCallParser.Create(toolDefinitions, enableThinking)is the same parser behindIToolCallParser.Qwen35ToolGrammar: the llguidance Lark grammar of the call format (experimental; see the runtime package for guided decoding).Qwen35ChatOptionsandQwen35Prompt: per-request settings (including the opt-inEscapeSpecialTokensandStopAtRoleMarker, which adds<|im_start|>to the stop token ids) and a rendered prompt.Qwen35Checkpoint: repository and revision, the special token ids (<|endoftext|>248044,<|im_start|>248045,<|im_end|>248046), the owned tokenizer profileqwen35-bpe-chatand its digest,Profile(aModelProfile: vocabulary 248,320, hidden size 1,024, 24 layers, head size 256, context 262,144, Apache-2.0) andLock.Qwen35Checkpoint.Lock: the asset lock of the Hub checkpoint itself (config.json, the safetensors weights and their index,tokenizer.json,tokenizer_config.json,chat_template.jinja; about 1.76 GB), for runtimes that load the original weights. The package also carries it aslocks/qwen3.5-0.8b.checkpoint.lock.json(JSON lock schema version 2, read byModelAssetLock.Load). Nothing downloads unless you pass it toModelAssetStore.DownloadAsync.
using Starlings.Generation;
using Starlings.Models.Qwen35;
using Starlings.Tokenizers;
string text = Qwen35ChatTemplate.Render([Qwen35Message.User("What is the capital of France?")]);
var tokenizer = Tokenizer.FromFile(tokenizerPath); // the qwen35-bpe-chat profile asset
int[] promptIds = [.. tokenizer.Encode(text, addSpecialTokens: false).Ids];
// ... run promptIds on any backend; stop at Qwen35Checkpoint.Profile.EosTokenIds ...
Qwen35ParsedOutput parsed = Qwen35ToolCallParser.Parse(completionText, tools: [], enableThinking: false);
Status of the definition surface added with the package split: Qwen35ChatTemplate.Instance and Qwen35ToolCallParser.Create (the IChatTemplate and IToolCallParser seams) are experimental: they give the same text and parts as the typed renderer and parser on every case of the template corpus, and no backend other than ONNX Runtime GenAI uses them yet. Qwen35Checkpoint.Lock is experimental: its sizes and digests were read from the Hub on 2026-10-26, and no runtime in this repository loads the checkpoint files yet. The Stable contract of the registry profiles below covers the ONNX Runtime GenAI runtime package; it is unchanged by the split.
The types of the runtime package (Qwen35ChatModel, Qwen35ModelAssets, Qwen35ModelOptions, Qwen35ModelInfo, Qwen35Capabilities) keep the namespace Starlings.Models.Qwen35.
Maturity
Maturity is per profile, generated from the registry manifests by model-pipeline catalog (do not edit this table by hand). Gates, evidence and capabilities per profile: the model catalog.
| Profile | Maturity | Platforms |
|---|---|---|
qwen3.5-0.8b/ortgenai-fp16x-cuda |
stable | linux-x64 cuda: supported; windows-x64 cuda: untested |
qwen3.5-0.8b/ortgenai-fp32x-cpu |
stable | linux-x64 cpu: supported; windows-x64 cpu: untested |
qwen3.5-0.8b/ortgenai-fp32x-cuda |
stable | linux-x64 cuda: supported; windows-x64 cuda: untested |
qwen3.5-0.8b/ortgenai-int4-cpu |
experimental | linux-x64 cpu: experimental; linux-x64 cuda: unsupported; windows-x64 cpu: untested |
The template renderer and the tool-call parser are part of every profile above; their statuses are in the runtime package's README and Qwen35Capabilities.
Attribution and license
Qwen3.5-0.8B by the Qwen team, Alibaba Cloud (Copyright 2026 Alibaba Cloud), Apache-2.0 (LICENSE in the Hub repository at the pinned commit). Qwen35ChatTemplate is a C# port of the checkpoint's chat_template.jinja (Apache-2.0); Qwen35ToolCallParser ports transformers' argument casting (Apache-2.0). This package's code is MIT OR Apache-2.0.
| Product | Versions Compatible and additional computed target framework versions. |
|---|---|
| .NET | net10.0 is compatible. net10.0-android was computed. net10.0-browser was computed. net10.0-ios was computed. net10.0-maccatalyst was computed. net10.0-macos was computed. net10.0-tvos was computed. net10.0-windows was computed. |
-
net10.0
- Starlings.Acquisition (= 0.5.0)
- Starlings.Generation.Abstractions (= 0.5.0)
- Starlings.Tokenizers (= 0.3.4)
NuGet packages (1)
Showing the top 1 NuGet packages that depend on Starlings.Models.Qwen35:
| Package | Downloads |
|---|---|
|
Starlings.Models.Qwen35.OnnxGenAI
Qwen/Qwen3.5-0.8B (revision 2fc06364) chat generation in-process on ONNX Runtime GenAI 0.17.1: Qwen35ChatModel over the model definition Starlings.Models.Qwen35 (the pinned chat template rendered in C#, the tool-call parser, prompt ids from the owned tokenizer Starlings.Tokenizers profile qwen35-bpe-chat: no Microsoft.ML.Tokenizers, no GenAI tokenizer) and Starling's decode loop (greedy or seeded sampling, stop tokens and strings, streaming, cancellation between steps, usage). Ships the asset lock of the ONNX Runtime GenAI bundle, not the weights. Profiles qwen3.5-0.8b/ortgenai-fp32x (CPU; CUDA with TF32 off) and ortgenai-fp16x (CUDA), Linux x64, Stable. The application adds Microsoft.ML.OnnxRuntimeGenAI or Microsoft.ML.OnnxRuntimeGenAI.Cuda. Maturity is per model profile (Stable, Preview or Experimental): see the README, generated from the model registry. |
GitHub repositories
This package is not used by any popular GitHub repositories.
| Version | Downloads | Last Updated |
|---|---|---|
| 0.5.0 | 197 | 11/3/2026 |
| 0.4.0 | 110 | 11/3/2026 |
| 0.3.0 | 245 | 10/19/2026 |
| 0.2.0 | 356 | 10/17/2026 |
| 0.1.0-preview.6 | 108 | 10/14/2026 |