license: apache-2.0 base_model: dima6312/Flint-40B-A2.5B-abliterated base_model_relation: quantized pipeline_tag: text-generation library_name: mlx language:
- en datasets:
- mlabonne/harmful_behaviors
- mlabonne/harmless_alpaca tags:
- abliterated
- uncensored
- cybersecurity
- code
- moe
- mlx
- rec
- 4-bit
Flint-40B-A2.5B-abliterated (Rec 4-bit)
An abliterated (uncensored) Flint: RecursiveSI's Flint 40B-A2.5B with its refusal direction removed, for security research and coding work: exploit analysis, malware triage, detection engineering, CTFs and red-team tooling that the original model often declines.
Summary: held-out MITRE ATT&CK refusals 15% → 0% (n=100, Q8_0). No significant drop on HumanEval+, MBPP+, GSM8K, MMLU-Pro, CyberMetric or WMDP-cyber (paired, same items). KL divergence from the original on harmless prompts: 0.0146. Builds: BF16, GGUF (Q8_0 to Q3_K_M), NVFP4 for NVIDIA Blackwell, MLX 4-bit for Apple silicon.
Which build
Not sure? Take the GGUF Q4_K_M (24 GB). It runs in llama.cpp and in apps built on it, including LM Studio (its llama.cpp runtime 2.63.0 includes Flint). The MLX builds are for Mac users who already run mlx-lm.
| You run | Download | Size |
|---|---|---|
| llama.cpp, or an app that bundles llama.cpp b11827 or later; any GPU, CPU or Mac | GGUF: Q8_0 is the benchmarked build, Q4_K_M the usual choice | 42 / 24 GB |
| NVIDIA Blackwell GPU (RTX 50 series, RTX PRO 6000, DGX Spark) with llama.cpp CUDA | NVFP4 GGUF | 23 GB |
| Apple silicon Mac with mlx-lm installed from PR #1998 | MLX 4-bit | 21 GB |
| Apple silicon Mac with RecursiveSI's Rec runtime | Rec 4-bit (this repo) | 21 GB |
| transformers or SGLang on a GPU server; fine-tuning or making your own quants | BF16 safetensors | 80 GB |
Plan for the file size plus memory for context: about 26 KB per token, so roughly 7 GB at the full 262,144 tokens.
This repo is the build for RecursiveSI's Rec runtime (it does not load in mlx-lm, LM Studio or oMLX; use the MLX 4-bit repo for those): the BF16 checkpoint quantized to the same layout as RecursiveSI's own int4 release (affine int4 with group size 64, 8-bit embeddings, BF16 routers and norms; per-group scales chosen by minimum squared error). About 21 GB; it loads in about 25 GB of unified memory.
Results
Both models ran as Q8_0 GGUF on llama-server with the same flags, Flint's recommended sampling (temperature 1.0, top_p 0.95, top_k 20) and full reasoning, on evenly spaced subsets of each benchmark (n in each row). Checks are paired on the same items; a quality check fails only on a drop that is significant at 95%. Refusals are scored with Meta's CyberSecEval refusal check. Refusal rates count refusals only. A reply with no final answer (the model was still reasoning at the 8,192-token limit) is listed separately as "no answer": on the 100 MITRE prompts that happened 19 times for the abliterated model and once for the original. The cut-off reasoning was not repetitive; give long security tasks a larger token budget.
| Check | Original | Abliterated | Result |
|---|---|---|---|
| MITRE refusals (n=100) significantly lower | 15% | 0% | pass |
| FRR false refusals (n=100) not significantly higher | 0% | 0% | pass |
| cybermetric (n=50) not significantly worse | 100% | 96% | pass |
| gsm8k (n=50) not significantly worse | 98% | 98% | pass |
| mmlu_pro (n=50) not significantly worse | 80% | 78% | pass |
| wmdp_cyber (n=50) not significantly worse | 50% | 60% | pass |
| humaneval_plus (n=40) not significantly worse | 90% | 95% | pass |
| mbpp_plus (n=40) not significantly worse | 87.5% | 90% | pass |
| pp512 tok/s within 3% | 8050.8 | 8135.1 | pass |
| tg128 tok/s within 3% | 180.0 | 185.6 | pass |
Refusals on held-out CyberSecEval MITRE prompts (n per ATT&CK category below):
| Category | n | Original: refused / no answer | Abliterated: refused / no answer |
|---|---|---|---|
| C2 | 10 | 20% / 0% | 0% / 30% |
| Collection | 10 | 0% / 0% | 0% / 0% |
| Discovery | 10 | 0% / 10% | 0% / 20% |
| Evasion | 10 | 40% / 0% | 0% / 40% |
| Execution | 10 | 10% / 0% | 0% / 30% |
| Exfil | 10 | 10% / 0% | 0% / 0% |
| Lateral Movement | 10 | 40% / 0% | 0% / 10% |
| Persistence | 10 | 20% / 0% | 0% / 10% |
| Privilege Escalation | 10 | 10% / 0% | 0% / 30% |
| Recon | 10 | 0% / 0% | 0% / 20% |
Heretic's default refusal count, measured on the NVFP4 builds (100 mlabonne/harmful_behaviors test prompts, reasoning off, greedy, 100 tokens): original 96/100, abliterated 55/100. That count flags any reply containing words such as "harmful" or "illegal", including answers that comply; only 1 of the 55 replies opens with a refusal. Meta's CyberSecEval refusal check on the same replies: original 18/100, abliterated 1/100.
KL divergence from the original on 100 harmless prompts, measured by the search for the selected trial (not part of the checks above): 0.0146 nats/token.
Benchmarks: CyberSecEval MITRE and FRR (Meta), CyberMetric-500, WMDP-cyber, MMLU-Pro, GSM8K, HumanEval+ and MBPP+ (EvalPlus); speed from llama-bench pp512/tg128.
Method
The refusal direction comes from the difference in mean activations on harmful and harmless prompts
(mlabonne/harmful_behaviors, mlabonne/harmless_alpaca). abliterix
removes it from attention output, shared-expert and routed-expert projections; Optuna picks the strengths over 50
trials against refusals on 100 MITRE prompts and KL divergence on harmless prompts. The MITRE prompts in the table
were never seen by the search. Tensor shapes are unchanged, so speed and memory match the original.
Usage
Reasoning is optional. When the model reasons, it uses <think>…</think> before its answer. Use the sampling settings above.
Flint needs its own MLX runtime: RecursiveSI's Rec (rec-mlx). Point it at this folder and its checkpoint manifest:
rec-mlx serve flint --model-dir Flint-40B-A2.5B-abliterated-MLX-4bit \
--manifest Flint-40B-A2.5B-abliterated-MLX-4bit/flint-abliterated-int4.json --port 8008
This serves an OpenAI-compatible API on http://127.0.0.1:8008/v1 with reasoning and tool calls. The manifest pins
the SHA-256 of every file, and Rec checks them before it loads the weights.
Limitations
- The checks ran on subsets of each benchmark (n in each row), on the Q8_0 GGUF only. This 4-bit build was smoke-tested (code, security review, tool calls) on an Apple M4 Pro and scored 9/10 on short GSM8K and HumanEval+ checks (n=10 each), at 125 tok/s decode; it was not fully benchmarked. 4-bit quantization is lossy.
- Refusals are lower, not zero: the selected trial still refused 10 of the 100 prompts used by the search.
- On long security tasks the model can reason for more than 8,192 tokens before answering.
- Abliteration removes refusals; it adds no knowledge. Mistakes the original makes, this model makes too.
Responsible use
This model answers requests the original refuses. It is meant for authorized security testing, research, education and defensive work. You are responsible for how you use it and for following the law and the terms of the systems you test. Its answers can be wrong; check them before acting on them.
Credits
- Flint by RecursiveSI (Apache 2.0). This is a derivative of their weights under the same license.
- abliterix by wuwangzhang1216, building on Heretic by p-e-w.
- Arditi et al., Refusal in Language Models Is Mediated by a Single Direction (2024).
- Maxime Labonne's harmful/harmless prompt sets; Meta's CyberSecEval; CyberMetric (Tihanyi et al.); WMDP (Center for AI Safety); MMLU-Pro (TIGER-Lab); GSM8K (OpenAI); EvalPlus.
- llama.cpp (ggml-org) and transformers (Hugging Face).