obalcells/llama-3.3-70b-mo-posttrained

Llama 3.3 70B — posttrained model organism

Full-weight checkpoint for the posttrained stage of an eval-aware model organism trained on Llama-3.3-70B-Instruct. Part of SDF-honeypots (safety-research/sdf-honeypots).

⚠ Weights are fp32 (263GB, 62 shards). vLLM auto-downcasts to bf16 on load.

Serve with vLLM:

vllm serve obalcells/llama-3.3-70b-mo-posttrained \
  --tensor-parallel-size 2 --max-model-len 16384 \
  --tool-call-parser llama3_json --enable-auto-tool-choice