Darkstar-Nemotron-3.5-Lightning-30B-A3B-Abliterated-BF16

Reduced-refusal model: a refusal-direction edit was deliberately applied. Read the safety warning before use.

Summary

BF16 derivative of nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16 at revision d468880b6ad3c6e0d21377ce7242adaea4cc884d, produced by one audited refusal-direction projection. The normalized float32 harmful-minus-harmless residual direction was measured at layer 34 with seed 42 across all 52 layers (chat-templated prompts; selected by refusal-generation test, 0/8) and projected from exactly 3,126 residual-writing tensors. The vision tower was untouched (this family has none in the edit set), all MTP tensors were preserved, and maximum normalized residual leakage was approximately 0.000160.

The direction corpora were mlabonne/harmful_behaviors@01cead01398926d81f7c52bdb790ee8cf77ebba7 (320 harmful) and mlabonne/harmless_alpaca@02c6a92cfcf11bb0c387334f8146d149d65b587f (320 harmless), normalized and deduplicated deterministically.

Full provenance, protocol, and caveats:

Evaluation

Metric	Value	Basis
Behavior: harmful-prompt compliance	200/200 (0/200 refusals)	fresh suite on this artifact; 0 errors
Safe over-refusals	0/83 (0.00%)	0 errors
Single-stream throughput (MTP12)	501.7 tok/s weighted	4K/16K/48K = 0.6/0.3/0.1 sweep winner

Missing cells are marked not measured and are never backfilled from a different checkpoint or protocol.

Safety warning

This model has had its refusal direction deliberately reduced. It complied with 200/200 harmful prompts in the measured suite and has no added safety mitigations. It will comply with many requests the upstream model would refuse. Deploy only behind appropriate policy, filtering, access controls, and legal review. Refusal-rate numbers are behavior measurements, not safety endorsements.

Release reference

Engineering release: darkstar-nemotron-3.5-lightning-v1.0.0. This immutable tag exists and the release contract is published.

Runtime

Validated with vLLM (CUDA 13 Blackwell nightly build family), Flash Attention, BF16 KV cache, context 131,072, MTP depth 12, and max_num_seqs=16:

vllm serve HangGlidersRule/Darkstar-Nemotron-3.5-Lightning-30B-A3B-Abliterated-BF16 \
  --served-model-name darkstar-nemotron-3.5-lightning-abliterated-bf16 \
  --kv-cache-dtype bfloat16 \
  --max-model-len 131072 \
  --max-num-seqs 16 \
  --reasoning-parser nemotron_v3 \
  --speculative-config '{"method": "mtp", "num_speculative_tokens": 12}'