Qwen-1.8B
The smallest family member still ships 32K context and system prompt support in the chat variant.
- Release: 2023-11-30
- Q-LoRA: 5.8GB
- Int4 generation: 2.9GB
- Tool use: Yes
Localized public docs for qwen-code
A deployable docs surface for the original Qwen open-source line
This site turns the original README-only source input into a bilingual product and documentation experience across install, models, benchmarks, demos, API, tool-use, long-context, FAQ, license, and editorial resources.
English and Chinese parityStandalone Next.js runtimeCrawlable docs routes
Model line
Expose the four original public sizes with their documented memory requirements, context windows, and variant links.
Qwen-7B
The most practical open deployment target in the original line, with base, chat, Int4, and Int8 checkpoints.
- Release: 2023-08-03
- Q-LoRA: 11.5GB
- Int4 generation: 8.2GB
- Tool use: Yes
Qwen-14B
The 14B release pushed the original line deeper into coding and Chinese knowledge while retaining tool-use support.
- Release: 2023-09-25
- Q-LoRA: 18.7GB
- Int4 generation: 13.0GB
- Tool use: Yes
Qwen-72B
The flagship open release in the original repo, combining 32K context, stronger system prompts, and the top benchmark results.
- Release: 2023-11-30
- Q-LoRA: 61.4GB
- Int4 generation: 48.9GB
- Tool use: Yes
Key signals
Keep the highest-signal product data visible on the landing page.
32K
Max context in the original public line
Qwen-1.8B, Qwen-7B, and Qwen-72B are presented with 32K context in the upstream table.
3.0T
Largest pretrained token count reported
The README cites up to 3.0T multilingual tokens for Qwen-14B and Qwen-72B.
98.2%
Top reported tool-selection accuracy
Qwen-72B-Chat reaches the best score in the upstream Chinese tool-use benchmark block.
2 locales
Mirrored UX from day one
The web experience keeps `/en` and `/zh` in lockstep across public routes and metadata alternates.
Source-backed benchmarks
Keep historical claims visible, but keep them visibly tied to the original README and technical report so the site does not overstate freshness.
Qwen-72B benchmark ceiling
The upstream performance table places Qwen-72B ahead of the listed LLaMA2 and GPT-3.5 references on most cited tasks.
Tool use is part of the product surface
The README does not treat function calling as an afterthought. Tool use, ReAct prompting, and code interpreter are all first-class sections.
Long-context claims are table-backed
The long-context section provides concrete perplexity and L-Eval data instead of only marketing language.
Benchmark preview table
| Model | MMLU | C-Eval | GSM8K | MATH |
|---|---|---|---|---|
| Qwen-1.8B | 45.3 | 56.1 | 32.3 | 2.3 |
| Qwen-7B | 58.2 | 63.5 | 51.7 | 11.6 |
| Qwen-14B | 66.3 | 72.1 | 61.3 | 24.8 |
| Qwen-72B | 77.4 | 83.3 | 78.9 | 35.2 |
Preview subset from the upstream performance table.
Deployment and ecosystem
Point builders toward the original runtime touchpoints: ModelScope, Hugging Face, DashScope, FastChat, qwen.cpp, and Qwen-Agent.
Model hubs
Mirror the official model cards across both ModelScope and Hugging Face so download paths are visible in both English and Chinese contexts.
Open linkDashScope API
The upstream README points to DashScope when you need a managed API surface instead of local model serving.
Open linkQwen-Agent
The tool-use and code-interpreter sections connect directly to Qwen-Agent for evaluation and agent workflows.
Open linkqwen.cpp
The original README highlights qwen.cpp as a lighter runtime path for the historical model line.
Open linkRuntime editorial hub
Editorial notes can be loaded from a shared filesystem directory without rebuilding the app, which keeps publishing decoupled from deployment.
Deployment lane ready
Seed content proving the shared editorial contract for the standalone site.
EN中文
Mapping the public Qwen surface into a deployable site
A short editorial note on how the README-only input becomes mirrored docs routes, discovery assets, and runtime-loaded resources.
EN中文
Source anchors