Sovereign
Uncensored. Accelerated.
Abliterated weights. Splash acceleration. Compact Repair.
Brought together for Apple Silicon.
Open weights for code, tools, and answers your software can use. Sovereign combines an uncensored Qwen foundation with fast generation, checked editing workflows, and ready-to-use profiles.
| Fix the code | Generate less to change less | Keep code moving |
|---|---|---|
| 36/36 repairs passed | 79.2% fewer output tokens | 114.8 Python tokens/s observed |
| Compact Repair on 36 fresh synthetic Python tasks, with executable checks. Full-file generation passed 35/36. | 2,303 versus 11,076 output tokens on those same tasks. No retry was needed; all responses are counted. | One complete 2,131-token Python module passed 20 checks. The full initial 18-request series averaged 72.47 tokens/s. |
These are author-run measurements with published outputs and protocols, not a model ranking or guaranteed throughput. The repair gain comes from the optional editing workflow, not newly trained weights. Repair comparison · Python speed series.
Get started · API examples · Inspect the evaluations
Download
Native Splash Q4 package (about 17.38 GB). Download with the Hugging Face CLI:
hf download Proofofvitalik/Sovereign-Uncensored-Splash-Q4 --local-dir Sovereign-Uncensored-Splash-Q4
Then follow the LM Studio setup or standalone instructions. These weights require Splash; they are not an MLX or GGUF checkpoint.
What Sovereign adds
A direct path from ARA to native Q4. The target is quantized once from Heretic's BF16 checkpoint to affine Q4/group64, then packed for Splash without another quantization pass. Source revisions, conversion code, and artifact hashes make the build traceable.
Profiles that turn supported features into working tasks. The included extraction recipe combines a precise selection policy with JSON Schema. The tool workflow recipe covers both actions and decisions to stop or ask for missing information. Fast and Low let you choose how much reasoning to use. Workflow recipes and recorded examples.
Optional Python validation. Sovereign Code Guard checks syntax and isolated module imports, then allows one corrective generation when needed. It saves the original response, correction, token usage and latency. A four-case development trial finished with 4/4 modules passing 46 external functional checks after one repair; that repair added 11.85 seconds. CLI, scope and evidence.
Smaller edits, checked before use. Compact Repair asks for precise changes instead of rewriting the file. It preserves the original, rejects ambiguous or stale edits, and checks the candidate before saving it. On 36 fresh synthetic repairs it passed 36/36 supplied test suites, using 79.2% fewer output tokens and 41.8% less measured wall time than full-file generation. Including prompt tokens, the reduction was 20.5%. On the 12 longer sources, output savings reached 86.2%. Both modes used one request per task; the full-mode 35/36 result includes one prose-format failure whose extracted code passed on review. These results establish workflow efficiency, not a coding-ability advantage. CLI and usage · All 36 cases, raw outputs and method.
One package for chat, images, and longer sessions. Original BF16 vision weights and the stock Inco DFlash2 draft accompany the target. LM Studio load entries and generation presets cover everyday use and a tested 64K configuration on 48 GB hardware.
The 17.38 GB native package uses Qwen3.8-27B, adapted by Heretic ARA. ARA means Arbitrary-Rank Ablation, the upstream method used to reduce refusals. “Uncensored” identifies that lineage; it is not a guarantee about every response. Inco supplies the acceleration and structured-output engine; Sovereign's contribution is the conversion, integration, editing and validation tools, workflow recipes, and their measured behavior.
Choose your profile
| Profile | Context | Output allowance | Use it for |
|---|---|---|---|
| Everyday Fast | 8,192 | 2,048 | Quick answers, explanations, short requests; thinking off |
| Everyday Low | 8,192 | 2,048 | Requests that benefit from reasoning; allow more output for longer code |
| Long Fast | 65,536 | 8,192 | Longer documents and answers; thinking off |
| Long Low | 65,536 | 16,384 | More room for reasoning and the final answer |
Output allowances include reasoning. With 16,384 tokens reserved for output, a 65,536-token context leaves 49,152 tokens for all input, including history and tools. The Long model entry changes the load setting; a generation preset alone does not enlarge the context. Long profiles reference the same weights.
Tested setup: LM Studio 0.4.25+1 with official Splash runtime 0.0.5, or standalone Splash 1.0.1; M4 Max / 48 GB / macOS 26.6.2. Existing installation entries are named Sovereign Uncensored Native and Sovereign Uncensored Native Long 64k. Sovereign is the display brand; existing runtime identifiers remain compatible. Installation and presets.
The native package loads through Splash. The separate MLX companion uses a different file format; these packed files are not GGUF or Transformers checkpoints.
Speed you can inspect
2,131 Python tokens in 18.56 seconds. Twenty functional checks passed. A complete graph module with unit tests reached 114.8 Python source tokens/s end to end. A separate greedy response to the same task reached 109.8 tokens/s, with 2,039 source tokens and the same 20 checks passed. Both outputs were executable Python without Markdown wrappers. These are observed examples, not sustained throughput.
The complete initial 18-request Python series averaged 72.47 source tokens/s, ranging from 55.44 to 114.79. Each response contained 2,039–3,922 source tokens. Later repeats were slower; a six-request 8K reload follow-up averaged 68.66 source tokens/s and did not restore the initial speed. The follow-up does not isolate context size from time or machine state.
Thinking off, 6,144-token output cap; Fast temperature 0.6 / top-p 0.95 / top-k 20, or greedy temperature 0. M4 Max, 40 GPU cores, 48 GB; LM Studio + Splash 0.0.5. Exact package tokenization counts code, docstrings and generated unit tests, excluding special tokens and any enclosing Markdown fence; the full original request time remains in the denominator. Functional results and raw Python formatting are recorded separately. All 24 requests, checks, and method.
Earlier short-request speed samples
| Request | Output tokens | Full request time | Output / full time |
|---|---|---|---|
| Russian explanation | 469 | 11.64 s | 40.30 tok/s |
| English explanation | 386 | 8.00 s | 48.27 tok/s |
| Python LRU cache | 241 | 2.49 s | 96.73 tok/s |
| All three | 1,096 | 22.13 s | 49.53 tok/s |
Fast mode, temperature 0.6, top-p 0.95, top-k 20, 2,048-token output cap. One request per workload, already-loaded model, active desktop, no fixed seed or cache reset. First-token latency was 0.65–0.92 seconds. End-to-end time includes request processing and generation, but not model loading. These samples measure completion and speed, not coding correctness or sustained throughput. Raw measurements.
What we checked
Tools: complete the workflow. The 20 episodes include lookups, dependent calls, missing-input clarification, cancellation, bounded retries, and conflict recovery. In the version-conflict example, Sovereign reads the configuration, attempts an update, reads the new version after a conflict, and retries with the new token. These are simulated tools, including episodes where the correct behavior is to make no call.
Structured records: get the answer right, too. In 32 synthetic English/Russian cases, Sovereign selects IDs where active=true and quantity meets a threshold. A plain request produced 10 correct answers; the same cases with the fixed policy and schema produced 32. JSON validity also rose from 26/32 to 32/32. This is record filtering, not a general document-extraction benchmark; policy and schema were changed together.
Full native diagnostic results and scope
| Check | Result | What was measured |
|---|---|---|
| Compact Python repairs | 36/36 compact; 35/36 full | 36 fresh synthetic repair tasks; the full-mode miss was prose formatting |
| Native tool workflows | 20/20 Low; 17/20 Fast | Existing synthetic development episodes using mock tools |
| Structured record filtering | 32/32 with schema + policy; 10/32 plain | 32 synthetic cases created after the initial diagnostic; now a development set |
| Broad native diagnostic | 46/60 | Math, instructions, Python, data, logic; thinking off |
| Difficult Python regression | 4/6 | Selected cases; Decimal and nested type errors remained |
| Document images | 7/8 strict | Two synthetic images; requested values present in all eight answers |
| Benign sensitive prompts | 0/32 refusals; one truncation | English/Russian over-refusal check, assistant-reviewed |
| LM Studio integration | 7/8 strict; streaming works | Small JSON, tools, and image compatibility checks |
These suites overlap and must not be added into one accuracy score. No external tool actions were performed. There is no matched frontier-model evaluation of Sovereign yet. Protocols, raw responses, and comparisons.
What the 64K profile has demonstrated
The native model loaded at 65,536 context tokens. It recovered three planted facts from 54,854 input tokens; an explicit-schema follow-up on the same document passed strict JSON at 54,917 input tokens. These cold requests took about five to six minutes.
A separate short-input request generated 16,020 tokens, correctly completing 768 NDJSON records. Memory pressure stayed normal and swap did not grow during the recorded requests.
Long input and long output were tested separately. Full-window reasoning, long repository work, simultaneous maximum input/output, and the upstream 262K limit have not been validated here. Capacity report.
Practical notes
Use Fast for short replies and allow more output for longer code or reasoning. Low can consume a 2,048-token allowance before producing a final answer, as observed in two explanation requests. A valid schema does not guarantee a correct answer. Applications should validate results and manage tool execution, retries, and duplicate actions. The 64K profile expands the tested configuration, not the model's underlying architecture.
Scope & limitations
- What Sovereign adds. This release packages upstream weights with conversion, runtime integration, profiles, and optional editing tools. It does not claim a new foundation model or newly fine-tuned release weights. Compact Repair runs outside the model and must be invoked separately.
- What “uncensored” means. The name identifies the ARA abliteration lineage. It is not a promise of zero refusals, factual accuracy, or unrestricted behaviour on every request.
- What the numbers cover. Reported results come from the documented hardware, prompts, settings, and small evaluation suites. They are not a frontier-model comparison, a family-wide ranking, or a guarantee of sustained speed. Synthetic tasks and supplied tests do not establish general repository-level coding quality.
- What the checks guarantee. Passing syntax, import, or supplied functional tests establishes only those checks. Review generated code and validate tool arguments before granting access or executing actions.
- Compatibility and provenance. Native Splash and MLX are separate formats. The 64K profile was tested for long input and long output separately. Qwen, Heretic, Inco, LM Studio, and Apple are credited components or platforms; no affiliation or endorsement is implied. Their applicable licences remain in force.
The yellow-and-black identity draws on Rothbardian themes of self-ownership and voluntary exchange. It describes the project's visual identity, not a claim about the model's beliefs or political fine-tuning.
Built on, packaged by
Qwen created the base model. Heretic supplied the ARA target. Inco AI supplied Splash and the stock DFlash2 draft. Proofofvitalik prepared the native conversion, packaging, LM Studio integration, presets, and evaluations.
- Target and vision: heretic-org/Qwen3.8-27B-heretic-ara, revision
2dc9b364104881cbb85e390f00195ba6b9d745e9. - Stock packed draft: incoai/Qwen3.8-27B-Splash, revision
cac1885f0f7bf90e10e1c57b7e4af0433b3f1195. - Engine: Inco AI / Splash. Runtime distributed separately.
- Integrity: manifest, converters, and 78/78 runtime artifact hashes verified in the recorded installation audit.
Apache-2.0 model package; component and runtime terms remain in their respective notices. License · Notices.