Positron AI BF16 Checkpoint
This repository contains an unquantized bf16 checkpoint of openai/gpt-oss-20b published by Positron AI.
The upstream release of openai/gpt-oss-20b ships its weights in MXFP4; this checkpoint is that release dequantized to bf16 by Positron AI. No further quantization was applied.
Recommended Use
Use this artifact as a full-precision (bf16) reference or baseline for openai/gpt-oss-20b, for example when comparing quantized builds against an unquantized checkpoint. It is not a quantized build and carries no quantization configuration.
Artifact Summary
| Field | Value |
|---|---|
| Base model | openai/gpt-oss-20b |
| Published artifact | positron-ai/openai_gpt-oss-20b-hf-bf16 |
| Weight precision | bf16 (dequantized from upstream MXFP4) |
| Quantization | none |
| Release date | 2026-08-25 |
| License | apache-2.0 |
Validation
KL-divergence is not reported: it measures drift between a quantized artifact and its bf16 reference, and this artifact is the bf16 reference.
License
openai/gpt-oss-20b is released under the Apache License 2.0. The upstream LICENSE and USAGE_POLICY files are included in this repository verbatim, copied from openai/gpt-oss-20b at revision 6cee5e81ee83917806bbde320786a8fb61efebee.
Provenance
This artifact was produced by Positron AI from openai/gpt-oss-20b. The original model license and usage restrictions continue to apply.