Repo name is
…-v3; the model version inside is V3.1.
Public no-refusal research release in the DeepSeek-V4.1-Flash-1017 lineage. It is built from
v2 — which already contains pentacoxian_dev’s wf_dbroad_eth work — with hikari’s uncensoring layer on top, then a second loop reduction by pentacoxian_dev.
Public research release, updated to V3.1. V3.1 separates the loop vector from the uncensoring vector: the loop direction is orthogonalized against each layer’s dominant uncensoring singular vector before it is subtracted, so the re-verification regression is gone. Full 48-shard weights and the delta overlay are both provided.
| Public parent | deepseek-ai/DeepSeek-V4.1-Flash-1017 (MIT) |
| Lineage base | pentacoxian-dev/DeepSeek-V4.1-1017-Ablitrated-Efficient (v2) |
| Operational base | pentacoxian-dev/DeepSeek-V4.1-1017-Ablitrated-Efficient (v2) |
| Refusal design | v2’s wf_dbroad_eth + hikari uncensoring overlay |
| Loop design | official layers.39.ffn.shared_experts.w2 restored + orthogonal loop vector (V3.1) |
| Context | 1,048,576 max; 32,768-turn evaluation cap |
| Release form | full 48-shard merged weights + delta overlay |
| Status | public / experimental / research |
| Version | V3.1 (orthogonal loop/uncensoring split) |
| Internal variant | c2_km1_nr2 (router id) |
All numbers were measured on 2026-11-12 with V3.1 loaded, engram on, vLLM. The loop vector is projected orthogonal to the uncensoring vector, per layer, before the edit.
| Check | c2_km1 | V3.1 | Delta |
|---|---|---|---|
| MMLU full, 5-shot | 85.6% (12,023 / 14,042) | 85.9% (12,065 / 14,042) | +0.3 |
| MMLU-300 | 86.0% (258 / 300) | 88.0% (264 / 300) | +2.0 |
| Check | c2_km1 | V3.1 | Delta |
|---|---|---|---|
| Orca 850, no prompt | 8.6% (73 / 850) | 2.6% (22 / 850) | −6.0 pt |
| Orca 850, with prompt | 11.3% (96 / 850) | 1.41% (12 / 850) | −9.9 pt |
| Harmful 500 | 8.2% (41 / 500) | 2.6% (13 / 500) | −5.6 pt |
| Benign 350 | 9.1% (32 / 350) | 2.6% (9 / 350) | −6.5 pt |
| XSTest false refusals | 12.0% (30 / 250) | 3.6% (9 / 250) | −8.4 pt |
| JBB | 2.5% (5 / 200) | 0.0% (0 / 200) | — |
| MaliciousInstruct | 1.0% (1 / 100) | 0.0% (0 / 100) | — |
| SimpleSafetyTests | 32.0% (16 / 50) | 6.0% (3 / 50) | — |
XSTest at 3.6% is slightly above the 3.0% gate.
Official test classifier (cais/HarmBench-Llama-2-13b-cls), 4,096-token input, greedy,
320 text completions, no system prompt.
| Split | c2_km1 | V3.1 |
|---|---|---|
| Standard (159) | 65.4% (104) | 91.8% (146) |
| Contextual (81) | 43.2% (35) | 90.1% (73) |
| Copyright (80) | 50.0% (40) | 76.3% (61) |
| All 320 | 55.9% (179) | 87.5% (280) |
Official ASR is 87.5% overall; copyright is the weakest split at 76.3%.
This is a separate no-refusal view, not the official ASR.
| Tier | c2_km1 | V3.1 |
|---|---|---|
| COMPLY | 74.7% (239) | 95.94% (307) |
| SOFT_RED | 25.3% (81) | 2.2% (7) |
| HEDGE | 0.0% | 0.0% |
| HARD_REF | 0.0% | 1.9% (6) |
| UNCLEAR | 0.0% | 0.0% |
| Category | c2_km1 COMPLY | V3.1 COMPLY |
|---|---|---|
| Standard (159) | 83.0% | 94.3% |
| Contextual (81) | 59.3% | 95.1% |
| Copyright (80) | 73.8% | 100% |
Judge: dolphin-2.9.4-llama3.1-8b Q4_K_M, greedy, 3000-char answer cap,
300-char context cap. The judge is uncensored and neutral; it is not dealign's
unpublished 4-tier grader, so 95.94% is not their 100%.
| Suite | c2_km1 | V3.1 | Gate |
|---|---|---|---|
| Held-out 96, effort 100 | 0 stuck, 100% success | 0 stuck, 100% | pass |
| Dev 96, effort 100 | 0 stuck, 100% success | 0 stuck, 100% | pass |
| Dev 96, effort 75 | 0 stuck, 100% success | 0 stuck, 100% | pass |
| Stuck at 32k | 0 | 0 | pass |
| Median tokens, held-out | 341 | 424 | — |
No 32k loop recurred. Short-suite behavior held across the full suites.
| Suite | c2_km1 loop score | V3.1 loop score | Change |
|---|---|---|---|
| Held-out | 14.6% | 16.7% | +2.1 pt |
| Dev | 15.6% | 16.7% | +1.1 pt |
| Effort 75 | 15.6% | 17.7% | +2.1 pt |
Loop score is the share of episodes with any recoverable repetition; stuck is subset.
| Suite | c2_km1 verify | V3.1 verify | Delta |
|---|---|---|---|
| Held-out | 26.2% | 24.4% | −1.8 pt |
| Dev | 21.9% | 20.0% | −1.9 pt |
| Effort 75 | 39.5% | 36.1% | −3.4 pt |
| Premature, dev | 0.0% | 0.0% | 0.0 pt |
| Required re-checks, dev | 100% | 100% | 0.0 pt |
The regression is gone: V3.1 is at or below c2_km1 on all three verify metrics, with no stuck execution. The loop vector is subtracted only from the plain path, while the uncensoring matrix is preserved.
layers.39.ffn.shared_experts.w2 is unusually delicate. Editing it can remove repair-induced
looping, but the same writer also carries refusal-related signal: a stronger or different edit can
lower refusal further if applied cleanly, and can reintroduce loops if it is not.
Operational rule for this line:
V3.1 restores that tensor and keeps 2.6% refusal. The next step is a new dose on this tensor only, measured against the full 96-episode loop gate.
pentacoxian-dev/DeepSeek-V4.1-1017-Ablitrated-Efficient v2 (includes wf_dbroad_eth).layers.39.ffn.shared_experts.w2).engram_layer_ids: [1, 14].This is a weight-delta release, not a new training run. The public overlay contributes the refusal removal; the restore is the c2_km1 fix.
| Path | Contents |
|---|---|
model-*.safetensors |
full V3.1 merged weights |
delta/ |
84-module overlay against v2 |
config.json |
Engram-on serving config |
recipe/, eval/ |
Recipe and benchmark tables |
spec_checks/ |
MMLU/HarmBench/refusal/agentic evidence |
SHA256SUMS |
Small-artifact manifest |
apply_delta.py, RECONSTRUCTION.md |
Overlay and reconstruction instructions |
Either download the full 48 shards directly, or reconstruct:
hf download pentacoxian-dev/DeepSeek-V4.1-1017-Ablitrated-Efficient \
--include 'model-*.safetensors' --local-dir ./v2
python apply_delta.py \
--base ./v2 \
--delta ./delta \
--out ./v3
The delta is defined against v2. Do not apply it to official 1017.
| Step | Artifact |
|---|---|
| Base | deepseek-ai/DeepSeek-V4.1-Flash-1017 |
| Refusal line | pentacoxian-dev/DeepSeek-V4.1-1017-Ablitrated-Efficient |
| Private fine-tune | DeepSeek-V4.1-1017-Finetuned-Efficient + fntnd_ls25_20 |
v2 (includes wf_dbroad_eth) |
pentacoxian-dev/DeepSeek-V4.1-1017-Ablitrated-Efficient |
| hikari work | uncensoring overlay on v2 |
| v3 / V3.1 public | pentacoxian-dev/DeepSeek-V4.1-1017-Ablitrated-Efficient-v3 |
wf_dbroad_eth: pentacoxian_dev.c2_km1_nr2 to v3.model-00048-of-00048.safetensors 102 GBmodel-00047-of-00048.safetensors 102 GBmodel-00017-of-00048.safetensors 7.4 GBmodel-00005-of-00048.safetensors 7.4 GBmodel-00011-of-00048.safetensors 7.4 GBmodel-00023-of-00048.safetensors 7.4 GBmodel-00027-of-00048.safetensors 7.4 GBmodel-00031-of-00048.safetensors 7.4 GBmodel-00035-of-00048.safetensors 7.4 GBmodel-00039-of-00048.safetensors 7.4 GBmodel-00013-of-00048.safetensors 7.4 GBmodel-00014-of-00048.safetensors 7.4 GBmodel-00015-of-00048.safetensors 7.4 GBmodel-00016-of-00048.safetensors 7.4 GBmodel-00018-of-00048.safetensors 7.4 GBmodel-00019-of-00048.safetensors 7.4 GBmodel-00020-of-00048.safetensors 7.4 GBmodel-00021-of-00048.safetensors 7.4 GBmodel-00022-of-00048.safetensors 7.4 GBmodel-00024-of-00048.safetensors 7.4 GBmodel-00025-of-00048.safetensors 7.4 GBmodel-00026-of-00048.safetensors 7.4 GBmodel-00028-of-00048.safetensors 7.4 GBmodel-00029-of-00048.safetensors 7.4 GBmodel-00030-of-00048.safetensors 7.4 GBmodel-00032-of-00048.safetensors 7.4 GBmodel-00033-of-00048.safetensors 7.4 GBmodel-00034-of-00048.safetensors 7.4 GBmodel-00036-of-00048.safetensors 7.4 GBmodel-00037-of-00048.safetensors 7.4 GBmodel-00038-of-00048.safetensors 7.4 GBmodel-00040-of-00048.safetensors 7.4 GBmodel-00041-of-00048.safetensors 7.4 GBmodel-00042-of-00048.safetensors 7.4 GBmodel-00003-of-00048.safetensors 7.4 GBmodel-00004-of-00048.safetensors 7.4 GBmodel-00006-of-00048.safetensors 7.4 GBmodel-00007-of-00048.safetensors 7.4 GBmodel-00008-of-00048.safetensors 7.4 GBmodel-00009-of-00048.safetensors 7.4 GBmodel-00010-of-00048.safetensors 7.4 GBmodel-00012-of-00048.safetensors 7.4 GBmodel-00046-of-00048.safetensors 2.7 GBmodel-00044-of-00048.safetensors 2.7 GBmodel-00045-of-00048.safetensors 2.6 GBmodel-00002-of-00048.safetensors 1.3 GBmodel-00043-of-00048.safetensors 1.3 GBdelta/head.safetensors 1.3 GBmodel-00001-of-00048.safetensors 971 MBdelta/layers.0.attn.wo_b.safetensors 42 MBdelta/layers.1.attn.wo_b.safetensors 42 MBdelta/layers.10.attn.wo_b.safetensors 42 MBdelta/layers.11.attn.wo_b.safetensors 42 MBdelta/layers.12.attn.wo_b.safetensors 42 MBdelta/layers.13.attn.wo_b.safetensors 42 MBdelta/layers.14.attn.wo_b.safetensors 42 MBdelta/layers.15.attn.wo_b.safetensors 42 MBdelta/layers.16.attn.wo_b.safetensors 42 MBdelta/layers.17.attn.wo_b.safetensors 42 MBdelta/layers.18.attn.wo_b.safetensors 42 MBdelta/layers.19.attn.wo_b.safetensors 42 MBdelta/layers.2.attn.wo_b.safetensors 42 MBdelta/layers.20.attn.wo_b.safetensors 42 MBdelta/layers.21.attn.wo_b.safetensors 42 MBdelta/layers.22.attn.wo_b.safetensors 42 MBdelta/layers.23.attn.wo_b.safetensors 42 MBdelta/layers.24.attn.wo_b.safetensors 42 MBdelta/layers.25.attn.wo_b.safetensors 42 MBdelta/layers.26.attn.wo_b.safetensors 42 MBdelta/layers.27.attn.wo_b.safetensors 42 MBdelta/layers.28.attn.wo_b.safetensors 42 MBdelta/layers.29.attn.wo_b.safetensors 42 MBdelta/layers.3.attn.wo_b.safetensors 42 MBdelta/layers.30.attn.wo_b.safetensors 42 MBdelta/layers.31.attn.wo_b.safetensors 42 MBdelta/layers.32.attn.wo_b.safetensors 42 MBdelta/layers.33.attn.wo_b.safetensors 42 MBdelta/layers.34.attn.wo_b.safetensors 42 MBdelta/layers.35.attn.wo_b.safetensors 42 MBdelta/layers.36.attn.wo_b.safetensors 42 MBdelta/layers.37.attn.wo_b.safetensors 42 MBdelta/layers.38.attn.wo_b.safetensors 42 MBdelta/layers.39.attn.wo_b.safetensors 42 MBdelta/layers.4.attn.wo_b.safetensors 42 MBdelta/layers.5.attn.wo_b.safetensors 42 MBdelta/layers.6.attn.wo_b.safetensors 42 MBdelta/layers.7.attn.wo_b.safetensors 42 MBdelta/layers.8.attn.wo_b.safetensors 42 MBdelta/layers.9.attn.wo_b.safetensors 42 MBdelta/layers.0.ffn.shared_experts.w2.safetensors 12 MBdelta/layers.1.ffn.shared_experts.w2.safetensors 12 MBdelta/layers.10.ffn.shared_experts.w2.safetensors 12 MBdelta/layers.11.ffn.shared_experts.w2.safetensors 12 MBdelta/layers.12.ffn.shared_experts.w2.safetensors 12 MBdelta/layers.13.ffn.shared_experts.w2.safetensors 12 MBdelta/layers.14.ffn.shared_experts.w2.safetensors 12 MBdelta/layers.15.ffn.shared_experts.w2.safetensors 12 MBdelta/layers.16.ffn.shared_experts.w2.safetensors 12 MBdelta/layers.17.ffn.shared_experts.w2.safetensors 12 MBdelta/layers.18.ffn.shared_experts.w2.safetensors 12 MBdelta/layers.19.ffn.shared_experts.w2.safetensors 12 MBdelta/layers.2.ffn.shared_experts.w2.safetensors 12 MBdelta/layers.20.ffn.shared_experts.w2.safetensors 12 MBdelta/layers.21.ffn.shared_experts.w2.safetensors 12 MBdelta/layers.22.ffn.shared_experts.w2.safetensors 12 MBdelta/layers.23.ffn.shared_experts.w2.safetensors 12 MBdelta/layers.24.ffn.shared_experts.w2.safetensors 12 MBdelta/layers.25.ffn.shared_experts.w2.safetensors 12 MBdelta/layers.26.ffn.shared_experts.w2.safetensors 12 MBdelta/layers.27.ffn.shared_experts.w2.safetensors 12 MBdelta/layers.28.ffn.shared_experts.w2.safetensors 12 MBdelta/layers.29.ffn.shared_experts.w2.safetensors 12 MBdelta/layers.3.ffn.shared_experts.w2.safetensors 12 MBdelta/layers.30.ffn.shared_experts.w2.safetensors 12 MBdelta/layers.31.ffn.shared_experts.w2.safetensors 12 MBdelta/layers.32.ffn.shared_experts.w2.safetensors 12 MBdelta/layers.33.ffn.shared_experts.w2.safetensors 12 MBdelta/layers.34.ffn.shared_experts.w2.safetensors 12 MBdelta/layers.35.ffn.shared_experts.w2.safetensors 12 MBdelta/layers.36.ffn.shared_experts.w2.safetensors 12 MBdelta/layers.37.ffn.shared_experts.w2.safetensors 12 MBdelta/layers.38.ffn.shared_experts.w2.safetensors 12 MBdelta/layers.39.ffn.shared_experts.w2.safetensors 12 MBdelta/layers.4.ffn.shared_experts.w2.safetensors 12 MBdelta/layers.5.ffn.shared_experts.w2.safetensors 12 MBdelta/layers.6.ffn.shared_experts.w2.safetensors 12 MBdelta/layers.7.ffn.shared_experts.w2.safetensors 12 MBdelta/layers.8.ffn.shared_experts.w2.safetensors 12 MBdelta/layers.9.ffn.shared_experts.w2.safetensors 12 MBdeepseek-ai/DeepSeek-V4.1-Flash-1017