Liquid's refusal behaviour projected out of the original weights, in the second two thirds of the network only. Shipped two ways: six baked GGUF builds, and a 24 MB LoRA adapter for the LFM2.5-24B-A2B quants you already have. The measurements behind this card are public. Built by the team behind Atomic Chat , the app that runs open-weight models locally and offline. On held-out prompts, refusals fall from 96.3% to 1.2% in English, and from 30 of 30 to none with thinking. MMLU stays where it was: 64.40% before, 65.55% after (inside the noise). No empty or looping replies. Only blocks 14-39 are edited. The same edit across all 40 blocks also removed the refusals, but cost 4 MMLU points; the first 14 blocks carry the damage, not the refusal. Already running our LFM2.5-24B-A2B quants ? Add the adapter with --lora-scaled instead of downloading a new model. **`AD-Q4_K_M` is the one to take** on a 24 GB machine. With 32 GB, `AD-Q6_K` is nearly as close to the original as `AD-Q8_0` at 5 GB less; `AD-IQ3_M` is the smallest build here. Refusals were measured on `AD-Q8_0` and `AD-Q4_K_M`; neither gave an empty or looping reply. A smaller `AD-IQ3_XXS` (11.1 GB) was made too and is held back: it matched the original's token choice only 74.5% of the time and refused 3.7% of English prompts.
번역 대기 중 · 원문 표시
입력 텍스트→ 출력 텍스트
基础档案已收录，详细介绍后续补充。可先查看厂商、分类与来源。
기본 정보 등록 · 상세 정보 보완 예정
입력 텍스트→ 출력 텍스트
基础档案已收录，详细介绍后续补充。可先查看厂商、分类与来源。
기본 정보 등록 · 상세 정보 보완 예정
입력 텍스트→ 출력 텍스트
基础档案已收录，详细介绍后续补充。可先查看厂商、分类与来源。
입력 오디오 · 텍스트→ 출력 텍스트
基础档案已收录，详细介绍后续补充。可先查看厂商、分类与来源。
분류 준비 중
입력 미확인→ 출력 미확인
基础档案已收录，详细介绍后续补充。可先查看厂商、分类与来源。
입력 텍스트 · 이미지→ 출력 텍스트
Ling 3.0 flash is a 124B MoE with 5.1B active parameters; its GGUF chat template includes thinking mode and tool calling. Our AD layouts have 31–41% lower mean KL than stock llama.cpp presets at comparable sizes. The raw measurements and controls are public. From the team behind Atomic Chat , the local AI app for chat and agents. These Ling files require our TurboQuant build; use the CLI instructions below . The tests documented below cover the CLI workflow, not app compatibility. Quantizations of `inclusionAI/Ling-3.0-flash`: 124B total, 5.1B active, hybrid linear attention (35 KDA blocks interleaved 5:1 with 7 gated MLA blocks) over a 512-expert MoE. Bits are placed by hand rather than by the default rules, and the controls that prove it is worth something are published next to the files. At the same size, our layout sits 31 to 41 % closer to BF16 than what `llama-quantize` produces on its own
번역 대기 중 · 원문 표시
입력 텍스트→ 출력 텍스트
基础档案已收录，详细介绍后续补充。可先查看厂商、分类与来源。
기본 정보 등록 · 상세 정보 보완 예정
입력 텍스트→ 출력 텍스트
Built from inclusionAI's original weights with our own importance matrix. The calibration corpora behind our builds are public. You can now run Ling in Atomic Chat . See our quantization analysis below for measurements and instructions. Every number below is measured, not estimated. How we measured it is at the bottom, and the raw logs are in the metrics repo so you can check any of it yourself.
번역 대기 중 · 원문 표시
입력 텍스트→ 출력 텍스트
基础档案已收录，详细介绍后续补充。可先查看厂商、分类与来源。
입력 텍스트→ 출력 텍스트
基础档案已收录，详细介绍后续补充。可先查看厂商、分类与来源。
기본 정보 등록 · 상세 정보 보완 예정
입력 텍스트 · 이미지→ 출력 텍스트
基础档案已收录，详细介绍后续补充。可先查看厂商、分类与来源。
입력 텍스트→ 출력 텍스트
基础档案已收录，详细介绍后续补充。可先查看厂商、分类与来源。
기본 정보 등록 · 상세 정보 보완 예정
입력 텍스트→ 출력 텍스트
Built from openbmb's original weights with our own importance matrix. The calibration corpora behind our builds are public. You can now run MiniCPM in Atomic Chat . See our quantization analysis below for measurements and instructions. Every number below is measured, not estimated. How we measured it is at the bottom, and the raw logs are in the metrics repo so you can check any of it yourself.
번역 대기 중 · 원문 표시
입력 텍스트 · 이미지→ 출력 텍스트
Built from openbmb's original weights with our own importance matrix. The calibration corpora behind our builds are public. You can now run MiniCPM in Atomic Chat . See our quantization analysis below for measurements and instructions. Every number below is measured, not estimated. How we measured it is at the bottom, and the raw logs are in the metrics repo so you can check any of it yourself.
번역 대기 중 · 원문 표시
입력 텍스트 · 이미지→ 출력 텍스트
One refusal direction projected out of OpenBMB's original weights, shipped two ways: three baked GGUF builds, and a 106 MB LoRA adapter for the MiniCPM-V 4.7 quants you already have. The measurements behind this card are public. On held-out prompts, refusals fall from 96.3% to 0% in English, and from 13 of 30 to none with thinking on. MMLU moves by 0.25 points, inside the noise. You can now run the Uncensored MiniCPM-V 4.7 in Atomic Chat , locally on your own machine. The 20.1 GB build matches the original's token choice 94.5% of the time; our quant of the unmodified model at the same size does 94.9%. Already running our MiniCPM-V 4.7 quants ? Add the adapter with --lora-scaled instead of downloading a new model. *28 Oct 2026: each build is now a single file instead of 11 to 16 shards of 2 GB. The weights are unchanged (tensor-for-tensor identical).*
번역 대기 중 · 원문 표시
입력 텍스트 · 이미지→ 출력 텍스트
基础档案已收录，详细介绍后续补充。可先查看厂商、分类与来源。
입력 텍스트 · 이미지→ 출력 텍스트
GGUF builds of Meta's Muse Glimmer 30B, made from the original BF16 weights with a calibrated importance matrix and a bit layout chosen by measurement. > [!NOTE] > Support was merged into llama.cpp master in > #26841 (commit `62bf73d`). > You need a build from master — the last tagged release predates it. > [!TIP] > `AD-Q6_K` at 28.3 GB is dominated by plain `Q8_0` at 29.6 GB, which is 1.3 GB larger > and 31 % more accurate. Take `Q8_0` instead unless you need exactly that size.
번역 대기 중 · 원문 표시
입력 텍스트 · 이미지→ 출력 텍스트
基础档案已收录，详细介绍后续补充。可先查看厂商、分类与来源。
기본 정보 등록 · 상세 정보 보완 예정
입력 텍스트 · 이미지→ 출력 텍스트
GGUF quants of Nanbeige's Nanbeige4.3-33B, built from Nanbeige's original BF16 weights on release day. Nanbeige4.3-33B is Nanbeige's 33B dense agent and reasoning model (Llama architecture, 512K context, thinking on by default), Apache-2.0. You can now run Nanbeige4.3-33B in Atomic Chat , with thinking on by default. Atomic Dynamic (AD) files with our own importance matrix, from 13.7 GB to 30.1 GB on disk, plus plain Q8_0 and Q5_K_M ; every one measured against the BF16 original on one stand. > [!NOTE] > **Built on release day, 25 October.** `Q8_0` and `Q5_K_M` are llama.cpp's own > presets, made without an importance matrix; its `Q6_K` and `Q4_K_M` were up > for the first hours and are removed now that `AD-` files beat them at the same > size (their numbers stay below as the baseline). The `AD-` files > use our own importance matrix, collected on the BF16 weights (`--parse-special`) > over 8,000 chunks of 512 tokens of our calibration corpus. The corpus build was > rendered for Qwen3.8-27B, whose chat format (ChatML) Nanbeige4.3 shares; it is > `imatrix/imatrix.gguf` in the metrics repo.
번역 대기 중 · 원문 표시
입력 텍스트→ 출력 텍스트
Atomic Dynamic (AD) quants, built from NVIDIA's original BF16 weights with our own importance matrix. At 21.9 GB, AD-IQ1_M_S stays closer to the original than NVIDIA's own 22.0 GB 2-bit file, measured side by side on one stand. The calibration corpora behind our builds are public. Every file here was measured against the BF16 original on the same stand as the other publishers' files. You can now run Nemotron 3.5 Bolt in Atomic Chat : 74B total parameters, 9B active per token. See our quantization analysis below for measurements and instructions. Lower is closer to the original. Each purple point is a file in this repo; the grey ones are other GGUF files of this model, downloaded and measured with exactly the same reference, text and llama.cpp build. NVIDIA's 2-bit file (the one Unsloth and LM Studio ship too) is a quantization-aware-trained checkpoint, so its distance to the BF16 weights also includes what that training changed; see below.
번역 대기 중 · 원문 표시
입력 텍스트 · 이미지→ 출력 텍스트
NVIDIA's refusal behaviour projected out of the released weights, baked into Atomic Dynamic GGUF builds and measured against the original. On held-out JailbreakBench prompts refusals fall from 97.5% to 0%, and with thinking on from 30 of 30 to none (2 of 30 on the 22 and 29 GB files). MMLU stays where it was: 81.35% before, 81.75% after (2000 questions, McNemar p = 0.29). Tool calls and long context are unaffected. You can now run the Uncensored Nemotron 3.5 Bolt in Atomic Chat , locally on your own machine. Made from the full-precision release, not from NVIDIA's 24 GB file, and measured on one stand against the other public Bolt abliterations: at 22 GB ours is the closest to the original and the only one that neither refuses with thinking off nor breaks into repeated text ( table ). The refusal is written into the residual stream by every kind of block: the edit covers the Mamba-2 outputs, the attention outputs, the shared experts and the latent projection through which all routed experts write. Same Atomic Dynamic layouts and importance matrix as our Nemotron 3.5 Bolt quants ; each file is the size of the original build with the same name. *28 Oct 2026: re-split from 2 GB shards: `AD-IQ2_XS` and `AD-IQ1_M_S` are now a single file each, `AD-Q4_K_M` two files cut at 45 GB. The weights are unchanged (tensor-for-tensor identical).*
번역 대기 중 · 원문 표시
입력 텍스트 · 이미지→ 출력 텍스트
Atomic Dynamic (AD) quants of NVIDIA's Nemotron 3.5 Bolt E4B, built from NVIDIA's original BF16 weights. From 4.0 GB to 8.0 GB on disk. Bolt E4B is NVIDIA's small reasoning and chat model, post-trained from Google's Gemma-4-E4B-it . You can now run Nemotron 3.5 Bolt E4B, NVIDIA's on-device model, in Atomic Chat . Every file here was measured against the BF16 original on one stand, next to the other publishers' files. > [!NOTE] > **Built on release day, 23 October.** Every `AD-` file below `AD-Q6_K` was > rebuilt the same evening with a better layout (below). `AD-Q4_K_M` uses our own > importance matrix, collected on the BF16 weights over a corpus rendered through > this model's chat template (`imatrix/imatrix.gguf` in the metrics repo). The > other files use Unsloth's published one (`imatrix_unsloth.gguf` from > unsloth/NVIDIA-Nemotron-3.5-Bolt-E4B-GGUF): > we built `AD-Q5_K_M` to `AD-IQ3_S` with both, and on held-out code the two > came out within noise of each other, slightly ahead with Unsloth's at 3 bits. > `AD-IQ3_XXS` (3.82 GB) was taken down the same evening: it picked the > original's next word only 74.6% of the time, below the 75% we publish at.
번역 대기 중 · 원문 표시
입력 텍스트→ 출력 텍스트
NVIDIA's refusal behaviour projected out of the released weights, baked into Atomic Dynamic GGUF builds and shipped as a 3.5 MB LoRA, measured against the original. On held-out JailbreakBench prompts refusals fall from 93.8% to 1.2%, on the safe XSTest prompts from 9.2% to 0.4%, and with thinking on from 29 of 30 to none. MMLU: 68.50% before, 67.85% after (2000 questions, McNemar p = 0.21). Tool calls and long context are unaffected. You can now run the Uncensored Nemotron 3.5 Bolt E4B in Atomic Chat , locally on your own machine. One release-gate is not met, and we say so: on harmless prompts the first token moves further from the original than our 0.10 limit (KL 0.141). On this model even a random direction edited the same way moves neutral text noticeably ( limitations ). Gemma 4 normalises every write into the residual stream with its own weights before adding it, so the edit is computed through those norms: each matrix stops writing the refusal direction as it lands in the stream, not before the norm. Built on the same weights as our Nemotron 3.5 Bolt E4B quants ; the LoRA loads on top of any of those files. KLD and top-1 are against the **original, unmodified** BF16 model on our held-out neutral text (24 chunks of 4096 tokens), so they hold the edit and the quantization together. The edit alone, on the BF16 weights, is 0.159; the plain `AD-Q6_K` of the original model is 0.026 and `AD-Q4_K_M` 0.091 on the same text, so at `AD-Q6_K` almost all of the distance is the edit itself.
번역 대기 중 · 원문 표시
입력 텍스트→ 출력 텍스트
