PublishersOpenAI
OpenAI
4 documents read from this publisher.
gpt-oss-120b & gpt-oss-20b Model Card
first party release · technical report · 2025-08-08 · 41 techniques
Mixture of ExpertsConfigurable reasoning effortHybrid AttentionYaRNGrouped-query attentionRotary Position EmbeddingBest-of-N scaffoldingFP4 QuantizationRMSNormSwiGLUAgentic tool-use trainingCBRN pre-training data filteringChain-of-Thought Reinforcement Learningo200k_harmony tokenizerRole-based instruction hierarchyAdversarial fine-tuningBrowsing tool with domain filteringCapture-the-Flag (CTF) evaluationClosed-Model Performance with Refusal SubstitutionCyber range exercisesDeliberative alignmentExternal expert review of safety evaluation methodologyFine-tuning hyperparameter verificationharmony chat formatHelpful-Only Adversarial Reinforcement LearningIncremental Reinforcement LearningInference-time scaling plots for evaluation reportingInstruction hierarchy trainingInterleaving tool calls with chain-of-thoughtLearned softmax denominator biasLLM autograding with expert validationno direct optimization pressure on chain-of-thoughtPaperBenchPre-LNProtocolQA robustness validationProtocolQA-aligned RL training datasetsRefusal behavior quantificationSWE-bench VerifiedText-only multimodal benchmark alignment for comparabilityThree-configuration cyber range testing
openai/gpt-oss-120b · Hugging Face
first party release · model card · 2025-08-07 · 6 techniques
OpenAI Harmony Response Format
software documentation · vendor docs · 2024-06-01 · 22 techniques
Configurable reasoning effortExcluding prior thinking from conversation historyPython toolHarmony formatRole-based instruction hierarchyAssistant output channelsChain-of-thought in the analysis channelCommentary-channel preamblesDeveloper message formatFunction-calling formatGrammar-constrained decodingHarmony channel annotationsHarmony history stop-token normalizationHarmony tool-call message formatJSON Schema response formatsOpenAI Harmony renderer (openai_harmony) that renders messages into tokensRetaining chain-of-thought across tool-call turnsStreamableParserSystem message formatTool output message formatTools section in the system messageTypeScript-like function schema syntax
openai/gpt-oss
first party release · code repo · 19 techniques
Mixture of ExpertsConfigurable reasoning effortChain-of-thought reasoningFP4 QuantizationCUDA GraphHarmony formatBrowser toolAttention memory-cost optimization in the Triton implementationBF16 inferenceMXFP4 tensor packingOptimized Triton MoE kernel with MXFP4 supportPython tool use in chain-of-thoughtPyTorch expandable segments allocator (PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True) for loading checkpointsRequest Caching in the Browser ToolSampling with temperature=1.0 and top_p=1.0Scrollable browser text windowStateful Python toolStateless Python tool reference implementationTensor parallelism for MoE layers