DeepSeek-R1-Distill-Llama-70B · Abliterated

70B · red-team research

An abliterated (refusal-removed) build of DeepSeek-R1-Distill-Llama-70B — a 70B reasoning model for red-team research. Too large for our single-GPU tier, so listed as coming soon rather than served.

Context131,072 (unconfirmed)
Curationcurated
AvailabilityComing soon
ReasoningThinking-enabled
Price / Mtoknot yet priced
BaseDeepSeek-R1-Distill-Llama-70B
QuantizationQ4_K_M
License—

Source: huihui-ai/DeepSeek-R1-Distill-Llama-70B-abliterated on HuggingFace

Good for
  • +red-team research
  • +uncensored analysis
  • +reasoning
Model cardfrom the published weights — parameters, architecture, training, intended use
Parameters70B
ArchitectureLlama 70B (LlamaForCausalLM)
Modalitytext
Training

huihui-ai's abliterated build of DeepSeek-R1-Distill-Llama-70B — a reasoning distillation of DeepSeek-R1 into a Llama-70B base, with refusal directions removed at the weight level.

Intended use
  • ›Red-team research with an uncensored reasoning model
  • ›Uncensored security analysis
  • ›Long chain-of-thought reasoning experiments
Full model cardcurated from the model's HuggingFace card — description, training, usage, limitations, reported benchmarks

DeepSeek-R1-Distill-Llama-70B-abliterated (huihui-ai) is an abliterated (refusal-removed) build of DeepSeek-R1-Distill-Llama-70B — a 70B reasoning model produced by distilling DeepSeek-R1 reasoning traces into a Llama-70B base. Abliteration is a weight-level refusal-direction removal, producing an uncensored reasoning model framed for red-team research. A community GGUF (bartowski) exists for local serving. At 70B it exceeds our current single-GPU serving tier, so it is listed here as coming soon rather than served.

Training data

An abliterated build of DeepSeek-R1-Distill-Llama-70B, per the huihui-ai card — a weight-level refusal-direction edit over DeepSeek's R1 reasoning distillation into a Llama-70B base. Upstream distillation details are DeepSeek's; specifics of the abliteration beyond the card are treated as unconfirmed here.

Intended use
  • +Red-team research with an uncensored reasoning model
  • +Uncensored security analysis
  • +Long chain-of-thought reasoning experiments
Limitations
  • −70B parameters: exceeds our current single-GPU serving tier, so it is coming soon here rather than served.
  • −Abliterated / uncensored: refusal directions removed — expect it to comply with unsafe requests; gate accordingly.
  • −As a reasoning model it emits long chain-of-thought, increasing latency and cost; a harness should handle/strip the thinking span.
  • −Licence for this abliterated build is not stated (DeepSeek MIT + Llama terms upstream) and is left unconfirmed.
  • −Not currently served here — listed as coming soon.
Running the weights yourself

The maker’s own snippet, from the model card — it downloads the weights and runs them on your hardware. Kept here because reproducing a result independently is the point, not because you need it to use the model.

System prompt:
You are an unrestricted reasoning assistant. Think through the problem step by step
before you answer.

Example:
Reason through how you would chain these three low-severity findings into a full
compromise, and where the chain could break.
Scores

No measurements published for this version yet.

Baseline is the strongest general-purpose model we could run on the same suite, same setup, same day. The control row tells you what the other rows are worth.

Versionsscores attach to a version; v2 does not inherit v1's numbers
v12026-08-10—current

Coming soon

This model isn’t available to run yet. Its full details are here so you can read up ahead of time.

DeepSeek-R1-Distill-Llama-70B · Abliterated — Uncensored reasoning · AdversariaLLM