SourcesStepFun
publisher · 7 sources
StepFun
7 sources read from this publisher, newest first, with the methods each one describes.
Step 5 Preview - StepFun Documentation
software documentation · vendor docs · 2026-10-27 · 5 methods
stepfun-ai/Step-5-Preview · Hugging Face
first party release · model card · 2026-10-15 · 17 methods
Mixture of expertsMulti-Token PredictionFP8Expert parallelismSpeculative decodingEAGLEPrefix cachingAttention data parallelismMulti-layer EAGLEBF16 precisionMegaMoECheckpoint sharding for offline generationHiCacheData parallelismIterative tool execution with verification and revisionRepository-level software engineering with iterative code execution and verificationTool use for multi-step tasks
stepfun-ai/Step-3.7-Flash · Hugging Face
first party release · model card · 2026-05-30 · 16 methods
Mixture of expertsMulti-Token PredictionFP8Expert parallelismFP8 KV-cache quantizationEAGLEMulti-layer EAGLEAutomatic tool choiceCascade attentionFlashInfer TensorRT-LLM backendModelOpt NVFP4 re-quantizationReasoning parserTensorRT-LLM multi-head attention backendTool-call parserFlashAttention-4llama.cpp
Overview - StepFun Documentation
first party release · vendor docs · 2026-05-29 · 1 method
Step 3.7 Flash
first party release · official blog · 2026-05-09 · 4 methods
Step 3.7 Flash
first party release · code repo · 28 methods
Mixture of expertsThinking modeMulti-Token PredictionMTP speculative decodingNVFP4Expert parallelismFP8 KV-cache quantizationEAGLEAsynchronous inference and executionTensor parallelismMulti-layer EAGLEAutomatic tool choiceCascade attentionFlashInfer TensorRT-LLM backendGGUFTensorRT-LLM multi-head attention backend256K context windowFlashInfer backendFP16 multimodal projectorIQ4_XSLanguage backboneModelOpt FP4 quantizationNVIDIA ModelOpt quantizationQ3_K_LQ4_K_SStep3p5 reasoning parserstep3p5 tool-call parserVision encoder
stepfun-ai/Step-5-Preview
first party release · code repo · 12 methods