Dokumentationerreichbar
OpenAI, README von gpt-oss beim Commit 5681a6b
Die README des Repos openai/
geprüft 10.11.2026
Worauf sich diese Seite beruft, wörtlich, abgerufen am 20.10.2026:
MXFP4 quantization: The models were post-trained with MXFP4 quantization of the Mo
bestätigt 10.11.2026E weights, making gpt-oss-120b run on a single 80GB GPU (like NVIDIA H100 or AMD MI300X) and the gpt-oss-20b model run within 16GB of memory. a non-optimized Py
bestätigt 10.11.2026Torch implementation for educational purposes only Requires at least 4× H100 GPUs due to lack of optimization.
bestätigt 10.11.2026In this implementation, we upcast all weights to BF16 and run the model in BF16.
bestätigt 10.11.2026This version can be run on a single 80GB GPU for gpt-oss-120b.
bestätigt 10.11.2026Both models were trained using our harmony response format and should only be used with this format; otherwise, they will not work correctly.
bestätigt 10.11.2026We released the models with native quantization support.
bestätigt 10.11.2026This will work with any chat completions-API compatible server listening on port 11434, like ollama.
bestätigt 10.11.2026These implementations are largely reference implementations for educational purposes and are not expected to be run in production.
bestätigt 10.11.2026so that the larger model can run with this code (e.g., on 4x
bestätigt 10.11.2026H100 or 2x H200) If you use Transformers' chat template, it will automatically apply the harmony response format.
bestätigt 10.11.2026uses an optimized triton Mo
bestätigt 10.11.2026E kernel that supports MXFP4
Laufzeitumgebung im Glossar05 Womit ein Modell läuft06 Vier Wege zum selben Modell