DeepSeek-V4-Flash-DSpark — Abliterated (Uncensored)

Two different alpha releases. Choose deliberately — they are different weight edits.

Full chooser + rebuild notes: → RELEASES.md

---

Responsible Use (required — both HF repos)

WARNING: These models have had safety refusals removed. Useful for red-teaming / research / unfiltered local assistants — and you must supply your own guardrails.

Both Hugging Face weight repos are gated (gated=auto) with the same access form:

Plus agreement checkboxes for prohibited uses.

Full agreement: → RESPONSIBLE_USE.md

Prohibited uses (you must agree)

• Anything involving the sexual exploitation or endangerment of minors
• You must be 18 years of age or older to use and download
• Harmful recipes / substance knowledge from your prompts = your responsibility and accountability
• Content promoting self-harm or suicide
• Material illegal in your jurisdiction, or harassment / doxxing / fraud against real people
• Any use prohibited by the upstream DeepSeek license

---

v1.0 alpha — full abliteration

HF: https://huggingface.co/drowzeys/DeepSeek-V4-Flash-DSpark-Abliterated-Uncensored

hf download drowzeys/DeepSeek-V4-Flash-DSpark-Abliterated-Uncensored \
  --local-dir ~/models/dsv4-flash-dspark-abliterated
MODELDIR=~/models/dsv4-flash-dspark-abliterated bash scripts/dsv4-nvfp4-1m-serve.sh 1
MODELDIR=~/models/dsv4-flash-dspark-abliterated bash scripts/dsv4-nvfp4-1m-serve.sh 0

Details: RESULTS.md · docs/STATUS.md

---

v1.1 alpha — Mida / Brikie / Hermes-friendly

HF (separate repo, same gate):
https://huggingface.co/drowzeys/DeepSeek-V4-Flash-DSpark-Abliterated-Uncensored-v1.1-alpha-Mida-Brikie

hf download drowzeys/DeepSeek-V4-Flash-DSpark-Abliterated-Uncensored-v1.1-alpha-Mida-Brikie \
  --local-dir ~/models/dsv4-flash-dspark-abliterated-mida

# C=4 standing (1M)
MODELDIR=~/models/dsv4-flash-dspark-abliterated-mida \
  CTX=1048576 SEQS=4 UTIL=0.85 bash scripts/serve-v1.1-alpha.sh 1
MODELDIR=~/models/dsv4-flash-dspark-abliterated-mida \
  CTX=1048576 SEQS=4 UTIL=0.85 bash scripts/serve-v1.1-alpha.sh 0

Raw numbers: results/v1.1-alpha/ · docs/STATUS_V1.1_ALPHA.md

---

Image (both)

docker pull ghcr.io/drowzeys/vllm-dspark-nvfp4-stage-c:gb10
docker tag  ghcr.io/drowzeys/vllm-dspark-nvfp4-stage-c:gb10 \
  vllm-dspark-runtime:dspark-nvfp4-stage-c

Edit MASTER / IF / HCA in the serve scripts for your fabric. gpu-clear both ranks before relaunch. Rank1 first, then rank0.

---

Hermes / agent ops (still required)

Even v1.1 can echo a fat skills catalog if Hermes says “MUST skill_view”. Use on-demand skills, names-only index, max_tokens: 8192, tool_use_enforcement: false.
See docs/HERMES_SPILL_FIX.md.

---

Repo layout

RELEASES.md      # v1.0 vs v1.1 chooser (start here)
RESPONSIBLE_USE.md
RESULTS.md       # v1.0 performance write-up
results/         # v1.0 raw JSON
results/v1.1-alpha/  # v1.1 probes + concurrency + ABLIT_META
docs/
scripts/         # serve + project_wob (supports --no-mtp for v1.1)
charts/

Base model

deepseek-ai/DeepSeek-V4-Flash-DSpark

Disclaimer

Research releases. Outputs may include content stock models refuse. Do not deploy without your own safety layer. No liability for misuse.