narendra747/gpt-oss-20b-nvfp4

GPT-OSS 20B NVFP4

This is the deterministic NML NVFP4 recipe-v1 conversion of unsloth/gpt-oss-20b-BF16@cc89b3e7fd423253264883a80a4fa5abc619649f. It uses last-axis one-dimensional blocks of 16 weights, low-nibble-first E2M1 payloads, positive E4M3FN block scales, and one F32 global factor per quantized parameter. Small and sensitivity-critical tensors remain BF16 exactly as declared by nml-source-tensors.json.

This artifact is intended for NML and is not mislabeled as the original GPT-OSS MXFP4 representation. Exact source hashes, tensor disposition, and conversion semantics are included in the repository. Recipe: nml-nvfp4-weight-v1.