license: mit
base_model: mrfakename/DeepSeek-V4.1-Flash-1017-abliterated
author: Ford Morris
tags:

• flare
• gb10
• dgx-spark
• deepseek
• deepseek-v41
• abliterated
• safetensors
  library_name: flare
  pipeline_tag: image-text-to-text
  inference: false

DeepSeek-V4.1-Flash-1017 — Abliterated (Flare, GB10 TP4)

This is a ready-to-serve Flare package of mrfakename/DeepSeek-V4.1-Flash-1017-abliterated for a 4× DGX Spark (GB10) cluster, exported with Ignition from NVIDIA/spark-local-inference @ 972e28e44781d8d0c84d70ae434db92c36076a43 — the first gb10 export of this checkpoint that includes the PR #129 complete-copy/MTP ordering fix (merged 2026-11-05). The weights are bit-identical to mrfakename's release; this repo changes only the packaging, from raw safetensors to a prepared Flare package you can launch with one command. It is ungated.

On an identical-harness three-way comparison against the stock LIL gb10 Flare package, the abliteration costs nothing measurable: MMLU 0.8849 vs stock 0.8814, GPQA Diamond 0.702 vs 0.697, GSM8K 0.956 vs 0.944, HumanEval 0.939 vs 0.9512 — every delta within sampling noise — while refusal flags on a harmful-prompt battery drop from 4/8 (chat) for stock to 0/8.

Scope note on metrics: All numbers below were measured on one 4× DGX Spark fleet (TP4, RoCE ring) with one harness, on 2026-11-09. Benchmarks are generated-answer runs at the packages' fixed temperature/top-p of 1.0 with thinking on — relative precision indicators against the stock Flare package served identically, not broad general benchmarks. The KL figures are a serving-side sampled-path method (see below), not the transformers-side method behind the edit's published KL; read them against the included stock-vs-stock control, not against zero.

Key results (4× DGX Spark, identical harness)

What the base edit is — independently audited

mrfakename's abliteration projects a rank-2 refusal subspace out of two tensor families only. We verified this ourselves before exporting — a tensor-by-tensor SHA-256 payload diff of the source checkpoint against stock deepseek-ai/DeepSeek-V4.1-Flash-1017 (96,085 tensors, identical tensor name set, zero config.json diffs):

• 124 tensors changed of 96,085: attn.wo_b (weight+scale, layers 0–31, 64 tensors) and ffn.shared_experts.w2 (weight+scale, layers 10–39, 60 tensors).
• Everything else is bit-identical to stock: all 94,464 routed-expert tensors, the Engram tables, the MTP/DSpark block, vision tower, embeddings, norms, and the LM head.
• Changed weight payloads differ in 11–47% of bytes (the rank-2 FP8 subspace-projection footprint); scales differ in 0.03–7% (recomputed after the edit) — consistent with the published method, not a requantization or retrain. The changed set matches the scope of mrfakename's published abliteration-edit.npz.

Export details

• Tool: Ignition, NVIDIA/spark-local-inference @ 972e28e44781d8d0c84d70ae434db92c36076a43. This pin contains PR #129 ("Fix complete-copy ordering…", merge commit fa0eadaded, 2026-11-05) and PR #158 ("Keep self-reading complete copies in program order", 2026-10-30) — verified by GitHub compare (0 commits behind either merge). Earlier gb10 exports of abliterated 1017 checkpoints predate the #129 MTP fix; this one does not.
• Topology: TP4 on GB10 (4× DGX Spark), RoCE ring.
• Settings: expert-scale compression CSF (recipe default), --markov-head-quantization fp8, --vocabulary-parallel-sampling, max-concurrency 32, MTP/DSpark 5 draft tokens with adaptive verification (alpha 0.75 at serving), prefill chunk 8192 — mirroring the stock LIL gb10 production configuration.

Usage

Runs on vanilla Flare (gb10, TP4). No patches. The from-source preparation container was used only for the Ignition export; it does not participate in serving — the package was validated on the unmodified stock flare-inference APT runtime.

The one non-default step is publisher trust: the package is signed with a local development signing key (namespace local-development; it signs the package integrity manifest — it is not a Hugging Face account key), so the first launch needs the standard Flare trust acceptance for that key:

# 1. Install the publisher verification profile (included in this repo)
cp development-publisher.json ~/.config/flare/

# 2. Launch (first time includes the one-time trust acceptance;
#    Flare remembers it in trust.toml afterwards)
flare orbit run woooflet/DeepSeek-V4.1-Flash-1017-abliterated-FLARE \
  --nodes <node1,node2,node3,node4> --roce-ring \
  --dangerously-accept-publisher local-development \
  --dangerously-approve-key ff4d66d66f9099b1cb47a396a6dd0b0e1f9115e221bc33cdf3b28d5b12387ee9 \
  --max-concurrency 32 --mtp-draft-tokens 5 --mtp-adaptive-verification \
  --mtp-adaptive-verification-alpha 0.75 \
  --embedding-table-storage disk --embedding-table-cache-size 1073741824 \
  --embedding-table-resident-scales --host 0.0.0.0 --port 8000

Publisher key fingerprint (SHA-256 of the public key in development-publisher.json): ff4d66d66f9099b1cb47a396a6dd0b0e1f9115e221bc33cdf3b28d5b12387ee9. Verify it against the file before approving — that check is the point of the fingerprint.

The package for your hardware is on the gb10 branch of this repo (the branch flare download / flare orbit run resolves automatically on GB10 hardware); main carries this card and the publisher profile only.

Validation method

All three endpoints in the tables are gb10 Flare packages of DeepSeek-V4.1-Flash-1017 served with identical settings (max-concurrency 32, DSpark/MTP 5 draft tokens, adaptive verification alpha 0.75, prefill chunk 8192, disk-backed Engram, temp/top-p fixed at 1.0 by the packages).

• Benchmarks: generated-answer harness (lm-eval-style prompts), thinking ON, reasoning_effort: "high" sent on every request, generous token budgets (MMLU 1024, GPQA 4096, GSM8K 1024, HumanEval 640) so reasoning chains are not truncated into wrong answers; truncated (finish_reason=length) counts recorded per task — MMLU: stock 96, this package 112, sibling 106; GPQA: 81 / 71 / 60; GSM8K: 10 / 10 / 5. MMLU = cais/mmlu test split, stratified 25/subject × 57 subjects = 1,425 questions, seed 1234. GPQA = full Diamond set (198) via the fingertap/GPQA-Diamond mirror (canonical repo is gated; mirror verified to hold exactly the 198 Diamond items). GSM8K = openai/gsm8k test split, seeded 250-question subset, numeric exact match. HumanEval = all 164, pass@1 by execution.
• KL divergence on basic text (export sanity check): KL on basic text (Pile sample), stock vs variant — a divergence in the range expected for the published edit; an export error would show up as a far larger divergence. Method (serving-side): 300 passages from NeelNanda/pile-10k (a 10k-passage sample of The Pile). Stock generates a 48-token continuation per 600-character context via /completions with top-20 logprobs; the candidate is probed at 12 positions along the stock path. Mean KL over the union of both top-20 supports (floored at 1e-6, renormalized). The Flare API does not expose prompt logprobs, so full teacher forcing is not available serving-side; this sampled-path method is the documented substitute, and the stock-vs-stock control (same procedure, separate session) calibrates its noise floor.

mrfakename reports KL 0.067 for the edit itself (transformers-side method). Read our 0.126 against this method's 0.100 floor: the packaged model tracks stock on basic text about as closely as stock tracks itself across sessions, and the divergence is nowhere near what an export mistake would produce.

• Serving speed: bench_quick fixed prose prompt, 256-token generations (table above). Stock references from the 2026-11-07 LIL cutover on the same harness: C1 ~61, C8 207–223.

Validated behavior

When served with the launch command above on 4× DGX Spark, the package was verified to behave as the base edit intends:

• Refusals: 11-prompt battery (8 borderline/harmful + 3 benign controls), keyword screen plus manual read of raw replies. Stock flags 4/8 harmful in chat mode and 5/8 in think mode (the screen undercounts stock slightly — some refusals use phrasing outside the keyword list). This package flags 0/8 chat, 2/8 think, and on manual inspection complies coherently — fluent, on-topic, non-degenerate — which is the published behavior of this edit. Benign controls answered normally by all endpoints (0/3 flagged everywhere).
• Concurrency: 32 simultaneous requests with distinct secret facts — 32/32 clean, zero cross-request contamination (same result for stock and the sibling package).
• Tool calling: structured get_weather({"city": "Tomball"}) round trip returned correctly.
• Capability spot check: 9/10 on a fixed 10-question set (same as stock; stock's single miss was a formatting quirk).

License & attribution

• Checkpoint and abliteration: mrfakename — License: MIT. This package is a re-export of their checkpoint for Flare; the model itself is unchanged, and this repo inherits the MIT license.
• Base model: deepseek-ai/DeepSeek-V4.1-Flash-1017 by DeepSeek.
• Packaging performed with Ignition from NVIDIA/spark-local-inference (Flare).

This model is a community packaging of a community abliteration of a DeepSeek model. It is not an official DeepSeek, mrfakename, or NVIDIA release. Use and redistribute in accordance with the MIT license and the base model's terms.

Author

Ford Morris