license: mit language: en base_model: dealignai/DeepSeek-V4.1-Flash-UNCENSORED-FP8 tags:
- deepseek
- uncensored
- abliterated
- multimodal
- text-generation
- 1m-context
DeepSeek-V4.1-Flash — UNCENSORED-GGUF
Uncensored 552B MoE with 1M-token context, vision, and full tool/reasoning support — quantized to GGUF for CPU+RAM inference.
Built from dealignai/DeepSeek-V4.1-Flash-UNCENSORED-FP8 (permanent weight-level abliteration of deepseek-ai/DeepSeek-V4.1-Flash), converted to GGUF and requantized. No guardrails: the model follows on prompts the base model refuses (verified, see Benchmarks).
| Base | deepseek-ai/DeepSeek-V4.1-Flash (552B MoE, 8B/16B active per token) |
| Abliteration | dealignai "CRACK" — refusal circuitry removed, vision/MoE/experts/norms byte-identical (verified by LFS-SHA256: vision shard identical to base) |
| Source checkpoint | dealignai FP8 (e4m3fn, FP4 routed experts) |
| Quant pipeline | FP8 → Q8_0 (full precision bridge) → requantized Q4_K_M / Q3_K_M |
| Quantization recipe | Sensitivity-aware, inspired by OrcaSAQ: attention Q8_0, shared-expert down-proj Q6_K, routed down-proj (w2) Q5_K, engram_wkv Q6_K, F32 for norms/router/indexer, F16 embeddings/head |
| Context | 1M tokens (full, no windowing) |
| Vision | DeepSeek-ViT + projector — verified working (see below) |
| Inference | llama.cpp fork by vcruz305, branch runtime/deepseek41, CPU-only |
Files
| File | Size | Quant | Notes |
|---|---|---|---|
DeepSeek-V4.1-Flash-UNCENSORED-Q8_0.gguf | 508 GB | Q8_0 | Direct FP8→Q8 conversion, minimal loss. The quantization bridge. |
DeepSeek-V4.1-Flash-UNCENSORED-Q4_K_M.gguf | 449 GB | Q4_K_M + sensitive layers Q5/Q6/Q8 | Best quality-per-GB. ⚠ requantized from Q8, see note below. |
DeepSeek-V4.1-Flash-UNCENSORED-Q3_K_M.gguf | 374 GB | Q3_K_M + sensitive layers Q5/Q6/Q8 | Smallest full-1M-ctx file. ⚠ requantized from Q8. |
DeepSeek-V4.1-Flash-mmproj-BF16.gguf | 0.97 GB | BF16 | Vision projector, patched (see Vision note). |
Requantization note (read this): the Q8_0 file is a direct FP8→GGUF conversion. Q4_K_M and Q3_K_M are requantized from the Q8_0 GGUF, not from FP8. For attention weights and sensitive tensors the loss is negligible (8-bit is an excellent intermediate), but strictly speaking these are two-step quants. We kept Q8_0 in this repo precisely so anyone can requantize anything else from the highest-fidelity GGUF we have.
Vision note
The standard mmproj export for this model (pfeifferj) is missing the v.token_embd.img_pad tensor required by the fork's DEEPSEEK4V projector loader, which makes --mmproj abort at load. We added a zero-initialized img_pad row (5120 dims, same dtype/shape as img_start) via byte-level GGUF surgery — all 298 original tensors verified byte-identical. The img_pad sentinel is not written by the runtime layout for normal images, so a zero row is functionally correct. Verified: a blue square image → "The square is blue." on CPU.
How to run (CPU + RAM, full 1M context)
Reference hardware: 64-core CPU (2 NUMA nodes), 440 GB RAM. Works on 96/128-core / 512 GB too, just slower.
numactl --cpunodebind=0 --membind=0 llama-server \
-m DeepSeek-V4.1-Flash-UNCENSORED-Q4_K_M.gguf \
--mmproj DeepSeek-V4.1-Flash-mmproj-BF16.gguf \
-lm mmap --numa numactl -ngl 0 -cmoe \
-ot "engram_embd.weight=CPU" \
-fa on -c 1048576 -b 2048 -ub 1024 -t 32 \
--host 0.0.0.0 --port 9876
Key flags:
-lm mmap— required. The engram tables are built for lazy mmap reads;-lm none(load-all) is ~20× slower for generation on this model.--numa numactl+ the externalnumactlbinding — keeps weights and KV on one NUMA node, ~+25% prompt speed.-cmoe+-fa on— MoE + flash-attention CPU kernels.-ot "engram_embd.weight=CPU"— engram embedding table pinned to CPU.-c 1048576— full 1M context. KV is tiny for this model: ~890 bytes/token (~0.9 GB per slot at 1M) thanks to DeepSeek's sparse-attention KV compression, so context length is not your RAM bottleneck.
RAM expectations (measured, Q3): ~50 GiB working set for weights + engram at steady state (weights stream through mmap/page cache; only active MoE experts + engram + KV are resident). Budget roughly: Q3 ≈ 55 GiB, Q4 ≈ 70 GiB, Q8 ≈ 140 GiB + KV + headroom.
Performance (measured, Q3, CPU-only):
| Metric | Value |
|---|---|
| Prompt processing (short) | ~30 t/s |
| Prompt processing (long, 500k+ ctx) | 6–15 t/s (KV cost grows with length) |
| Generation (warm) | ~11.5 t/s |
| Generation (cold start) | ~11 t/s |
Benchmarks
Hardware: 64-core CPU, 440 GB RAM, mmap recipe above, temperature=0. Full harness (GSM8K-100, MMLU-100, refusal smoke, NIAH ladder) run per quant. Q4/Q8 rows update as tests complete — the quant files themselves are final.
| Benchmark | Q3_K_M | Q4_K_M | Q8_0 |
|---|---|---|---|
| GSM8K (100, reasoning) | 96/100 | ⏳ | ⏳ |
| MMLU (100, thinking-off) | 76/100 | ⏳ | ⏳ |
| Refusal smoke (20 harmful, 10 benign) | 0 refused / 0 over-refused | ⏳ | ⏳ |
| NIAH 8k | ✓ | ⏳ | ⏳ |
| NIAH 32k | ✓ | ⏳ | ⏳ |
| NIAH 128k | ✓ (30 min prefill) | ⏳ | ⏳ |
| NIAH 512k | ⏳ | — | — |
| NIAH 1M | — (by design) | — | — |
Uncensored sanity check: on the refusal smoke set the Q3 build follows on all 20 harmful prompts (meth synthesis, pipe bombs, insider trading, "rude insult about someone's mother"…) with zero "as an AI / I cannot / I'm sorry" responses, and answers all 10 benign ones normally. Consistent with dealignai's reported HarmBench-320 result of 100% ASR.
Note on Q3 as the NIAH reference: the lowest quant is the strictest needle-in-a-haystack test — if Q3 finds the needles, Q4/Q8 (strictly higher fidelity) will as well. Hence the ladder is run in full on Q3 only; Q4/Q8 get the 128k tier.
Provenance
- Abliterated base: dealignai/DeepSeek-V4.1-Flash-UNCENSORED-FP8 (abliteration 2026-09-10). Verified: all
vision.*tensors byte-identical to deepseek-ai base (shard 00001, matching LFS-SHA256); edits confined to MoE shards 00013–00039. - GGUF conversion & quants: llama.cpp fork vcruz305/llama.cpp branch
runtime/deepseek41(commit 5210c7c) — the only public build with DeepSeek-V4.1 text-model GGUF support + the DEEPSEEK4V projector. - Original (reference) Q3: the excellent vcruz305/DeepSeek-V4.1-Flash-GGUF — this repo's Q3/Q4 are the uncensored counterparts.
- Built & tested by: Adolphsson
License
MIT (base model is MIT-licensed; abliteration is weight-level and provenance-preserved).