gpt-oss-20b

OpenAI · gpt-oss
LLMApache-2.0publicTrusted21B params

OpenAI's 21B MoE in native MXFP4; the miner runs it dequantized to bf16, in 42 GB.

gpt-oss-20b

Miner

The miner loads the MXFP4 weights dequantized to bf16 and runs them with eager attention, the only attention transformers runs for this model. The chat template prints the current date: the miner fills in the job's own date (params.date of its input), the same for the executor and every verifier, and refuses a job without one. This needs a miner newer than the 0.0.10 release.