How qwenlm/qwen3 works
Qwen3 is an open-weight language-model release repository whose main user workflow is to obtain a checkpoint and follow guidance to run inference. The README also presents local serving, large-scale deployment, quantization, post-training, and application-framework integration as distinct capabilities. The repository contains no model runtime implementation in the supplied tree; these capabilities are represented by their documentation paths. A separate sampled evaluation workflow generates answers through an OpenAI-compatible endpoint, saves JSONL output, then dispatches configured scoring. Relationships for documented capabilities use README evidence; source-level evaluation links are limited to the excerpts shown.
Main entry points: infer_multithread.py Answer generation · quickstart.md Quickstart demos · utils_vllm.py Chat API client · eval.py Evaluation runner
Components 19
Model access
- Qwen3 checkpoints
README.md
Inference and serving
- Quickstart demos
docs/source/getting_started/quickstart.md - Transformers inference
docs/source/inference/transformers.md - Local runtimes
docs/source/run_locally - Inference deployment
docs/source/deployment - Quantization guides
docs/source/quantization - Application frameworks
docs/source/framework
Model adaptation
- Post-training guides
docs/source/training
Evaluation tooling
- Performance benchmarks
examples/speed-benchmark - Answer generation
eval/generate_api_answers/infer_multithread.py - Chat API client
eval/generate_api_answers/utils_vllm.py - Generated answers
eval/output - Evaluation runner
eval/eval/eval.py - ARC-AGI scorer
eval/eval/arc_agi_1.py - Detailed scores
eval/eval_res
Other components
- Model user
- Model hubs
- Evaluation operator
- OpenAI-compatible server
Connections 17
- Model user → Model hubs: searches checkpoints
README.md - Model hubs → Qwen3 checkpoints: provides weights
README.md - Model user → Quickstart demos: follows guidance
README.md - Quickstart demos → Transformers inference: guides inference
README.md - Model user → Local runtimes: runs locally
README.md - Model user → Inference deployment: deploys at scale
README.md - Model user → Quantization guides: quantizes checkpoints
README.md - Model user → Post-training guides: post-trains models
README.md - Model user → Application frameworks: integrates applications
README.md - Model user → Performance benchmarks: checks performance
README.md - Evaluation operator → Answer generation: runs generation
eval/generate_api_answers/infer_multithread.py - Answer generation → Chat API client: requests completions
eval/generate_api_answers/infer_multithread.py - Chat API client → OpenAI-compatible server: calls chat API
eval/generate_api_answers/utils_vllm.py - Answer generation → Generated answers: appends generations
eval/generate_api_answers/infer_multithread.py - Evaluation operator → Evaluation runner: runs configured scoring
eval/eval/eval.py - Evaluation runner → ARC-AGI scorer: dispatches task scorer
eval/eval/eval.py - ARC-AGI scorer → Detailed scores: writes detailed scores
eval/eval/arc_agi_1.py