01
failspy
Source-based descriptionfailspy’s Codestral-22B-v0.1-abliterated-v3 is a derivative of mistralai/Codestral-22B-v0.1 whose bfloat16 safetensor weights were modified by orthogonalizing out the model’s strongest refusal directions. According to the model card, the creator says the aim is to inhibit refusal while leaving the model tuned in all other respects like the original, rather than applying a behavioral fine-tune. The creator cautions that refusal is not guaranteed to disappear: the model may still refuse requests, lecture about ethics or safety, or exhibit quirks from the orthogonalization. The repository is provided in safetensors format and its Hugging Face metadata lists the license as other.
Model type (creator-reported)abliterated
Size / main format22.2473Bsafetensors
02
oyildirim
Source-based descriptionCyberStrike-OffSec-35B is a Qwen3.6-35B-A3B fine-tune for authorized security-testing agents. Its changes target the CyberStrike harness's tool-call format and agent routing rather than adding security knowledge. The release includes a merged checkpoint, with an adapter on a separate revision. The author documents repetition-loop risks and stresses using the full tool schema.
Model type (creator-reported)fine-tuned
Size / main format35.1072BsafetensorsGGUF version available
03
dphn
Source-based descriptionA full-weight Yi-1.5-34B fine-tune from the Dolphin team, using ChatML for conversation. The creator describes a training mix that includes instruction following, coding and function-calling data, and characterizes the result as uncensored. Its card distinguishes 4k base positional embeddings from 8k training sequences; neither establishes a tested runtime context limit here. Safetensors weights and Apache-2.0 license metadata are available.
Model type (creator-reported)fine-tuned
Size / main format34.3889BsafetensorsGGUF version available
04
dphn
Source-based descriptionA Mistral Small 24B Instruct fine-tune made by Dolphin with Venice. The creators describe it as uncensored and emphasize setting your own system prompt. The card recommends a low temperature, such as 0.15. This repository provides the canonical safetensors checkpoint rather than an FP8 or GGUF package. The creator's setup advice and claims about refusal behavior have not been tested here.
Model type (creator-reported)fine-tuned
Size / main format24.0114BsafetensorsGGUF version available
05
dima6312
Source-based descriptiondima6312's abliterated derivative of RecursiveSI's Flint 40B-A2.5B, released as BF16 safetensors under Apache 2.0. The creator describes it as an uncensored version with the refusal direction removed for security research and coding tasks such as exploit analysis, malware triage, detection engineering, CTFs, and red-team tooling. According to the model card, the refusal direction is computed from mean activation differences on mlabonne/harmful_behaviors and mlabonne/harmless_alpaca, and abliterix removes it from attention output, shared-expert, and routed-expert projections. The creator reports held-out MITRE ATT&CK refusals dropped from 15% to 0% and no significant drop on HumanEval+, MBPP+, GSM8K, MMLU-Pro, CyberMetric, or WMDP-cyber in paired checks. The card also notes refusals are lower, not zero, and that abliteration adds no knowledge. This repository contains the BF16 safetensors weights; other quantized builds are separate.
Model type (creator-reported)abliterateduncensoredfine-tuned
Size / main format39.7853BsafetensorsGGUF version available
06
mlabonne
Source-based descriptionmlabonne's uncensored derivative of google/gemma-3-12b-it, provided as safetensors weights under the gemma license. The creator says it uses an improved abliteration technique that targets refusal directions in specific model modules, aiming for an acceptance rate above 90% while maintaining coherent outputs. According to the model card, the process involves orthogonalizing hidden states to subtract refusal behavior, with weight factors following a normal distribution. The card recommends generation parameters of temperature=1.0, top_k=64, and top_p=0.95. This is not a fine-tuned model; the creator describes it as an abliterated version.
Model type (creator-reported)abliteratedfine-tuned
Size / main format11.766Bsafetensors
07
mlabonne
Source-based descriptionmlabonne's refusal-modified derivative of google/gemma-3-12b-it-qat-q4_0-unquantized, released as safetensors weights under the gemma license. The creator describes it as an uncensored version made with a new abliteration technique that targets refusals with enhanced accuracy. According to the model card, the refusal direction is computed by comparing residual streams between harmful and harmless samples, and hidden states of target modules are orthogonalized to subtract that direction. The creator says a hybrid evaluation with a dedicated test set and NousResearch/Minos-v1 was used, with a goal of acceptance rate above 90% while retaining coherent outputs. These refusal and coherence claims have not been independently tested here.
Model type (creator-reported)abliteratedfine-tuned
Size / main format11.766Bsafetensors
08
mlabonne
Source-based descriptionmlabonne's experimental abliterated derivative of google/gemma-3-1b-it, released as safetensors weights under the Gemma license. The creator says it uses layer-wise abliteration on hidden states (layers 3–45) with a refusal weight of 0.75 to remove refusals while aiming to preserve capabilities, noting occasional garbled text. According to the model card, this method achieved over 90% acceptance rate in their tests. The supplied configuration sets the architecture to Gemma3ForCausalLM, which is a configuration value rather than a verified property of these modified weights. No independent evaluation of refusal behavior or output quality is provided here.
Model type (creator-reported)abliteratedfine-tuned
Size / main format999.886Msafetensors
09
mlabonne
Source-based descriptionmlabonne's abliterated derivative of google/gemma-3-1b-it, provided as safetensors weights under the gemma license. The creator describes it as an uncensored version made with a new abliteration technique and says this v2 targets refusals with enhanced accuracy. According to the model card, the method computes a refusal direction from harmful and harmless samples and orthogonalizes hidden states of target modules such as o_proj to subtract it. The creator recommends generation parameters temperature=1.0, top_k=64, top_p=0.95 and says the goal is an acceptance rate above 90% while still producing coherent outputs. These refusal and coherence claims have not been independently tested here.
Model type (creator-reported)abliteratedfine-tuned
Size / main format999.886Msafetensors
10
mlabonne
Source-based descriptionmlabonne's refusal-modified derivative of google/gemma-3-1b-it-qat-q4_0-unquantized, provided as safetensors weights under the gemma license. The model card describes it as an uncensored version created with a new abliteration technique that targets refusals with enhanced accuracy. According to the creator, the refusal direction is computed by comparing residual streams between harmful and harmless samples, and hidden states of target modules such as o_proj are orthogonalized to subtract that direction. The creator says this is an improved version and recommends generation parameters temperature=1.0, top_k=64, top_p=0.95. The card states the goal is an acceptance rate above 90% while still producing coherent outputs, but this has not been independently tested here.
Model type (creator-reported)abliteratedfine-tuned
Size / main format999.886Msafetensors
11
mlabonne
Source-based descriptionmlabonne's uncensored derivative of google/gemma-3-27b-it-qat-q4_0-unquantized, released as safetensors weights under the gemma license. The creator describes it as abliterated using a new technique that computes a refusal direction by comparing residual streams from harmful and harmless samples, then orthogonalizes hidden states of target modules such as o_proj to subtract that direction. According to the model card, this version targets refusals with enhanced accuracy and recommends generation parameters temperature=1.0, top_k=64, top_p=0.95. The creator says a hybrid evaluation with a dedicated test set and NousResearch/Minos-v1 was used to measure acceptance rate, with a goal above 90% while retaining coherent outputs. These claims have not been independently tested here.
Model type (creator-reported)abliteratedfine-tuned
Size / main format27.0093Bsafetensors
12
mlabonne
Source-based descriptionmlabonne's refusal-modified derivative of google/gemma-3-4b-it, provided as safetensors weights under the gemma license. The model card describes it as an uncensored version created with a new abliteration technique that targets refusals with enhanced accuracy. According to the creator, the refusal direction is computed by comparing residual streams between harmful and harmless samples, then hidden states of target modules such as o_proj are orthogonalized to subtract that direction. The creator says this version uses a hybrid evaluation with a dictionary approach and NousResearch/Minos-v1, aiming for an acceptance rate above 90% while still producing coherent outputs. The creator recommends generation parameters temperature=1.0, top_k=64, and top_p=0.95. Refusal behavior and coherence claims have not been independently tested here.
Model type (creator-reported)abliteratedfine-tuned
Size / main format3.88026Bsafetensors
13
mlabonne
Source-based descriptionmlabonne's abliterated derivative of google/gemma-3-4b-it-qat-q4_0-unquantized, provided as safetensors weights under the gemma license. The model card describes it as an uncensored version created with a new abliteration technique that computes a refusal direction by comparing residual streams from harmful and harmless samples, then orthogonalizes hidden states of target modules such as o_proj to subtract that direction. The creator says this improved version targets refusals with enhanced accuracy and recommends generation parameters temperature=1.0, top_k=64, and top_p=0.95. The metadata lists image-text-to-text as the pipeline. Refusal behavior and coherence claims have not been independently tested here.
Model type (creator-reported)abliteratedfine-tuned
Size / main format3.88026Bsafetensors
14
google
Source-based descriptionGemma 4 26B A4B IT is Google's instruction-tuned mixture-of-experts model for text and image conversations, coding and tool use. It activates a subset of its experts for each token, unlike the dense 31B edition. Google documents a 256K context window and switchable reasoning mode. This is the standard instruction-tuned release, with safety training rather than an uncensoring modification; native audio is limited to other Gemma 4 sizes.
Model type (creator-reported)fine-tuned
Size / main format25.8059BsafetensorsGGUF version available
15
google
Source-based descriptionThis is the unquantized, quantization-aware-trained version of Google's instruction-tuned Gemma 4 26B A4B mixture-of-experts model. Despite Q4_0 in the name, the repository contains half-precision weights extracted from the QAT pipeline for downstream compilation or research, not the ready-to-run four-bit GGUF edition. Google's card also says speculative-decoding assistant checkpoints must match the target's QAT precision.
Model type (creator-reported)fine-tuned
Size / main format25.8059BsafetensorsGGUF version available
16
Gryphe
Source-based descriptionA writing-style fine-tune of Gemma-4-26B-A4B-it. Gryphe says V2 trains only the lm_head output projection for one epoch using narrative data, rather than retraining every model layer. The creator's sampling example uses temperature 1.0, MinP 0.10 and DRY. This is the V2 safetensors checkpoint, separate from the 31B and 12B StyleTune releases. Claimed style improvements and preserved general capabilities have not been independently tested here.
Model type (creator-reported)fine-tuned
Size / main format26.5441BsafetensorsGGUF version available
17
google
Source-based descriptionGemma 4 31B IT is Google's dense instruction-tuned model for text and image input, with text output. It combines local and global attention, a documented 256K context window and optional reasoning, with native function calling for tool-based applications. Unlike the 26B A4B edition, it is not a mixture-of-experts model. The card's speech-processing features belong to the E2B, E4B and 12B editions, not this checkpoint.
Model type (creator-reported)fine-tuned
Size / main format31.2731BsafetensorsGGUF version available
18
llmfan46
Source-based descriptionllmfan46’s gemma-4-31B-it-uncensored-heretic is a derivative of Google’s gemma‑4‑31B‑it model. According to the model card, the creator applied the Heretic v1.2.0 tool with the Arbitrary‑Rank Ablation (ARA) method, ablating layers 30‑48 and targeting the attn.o_proj component to reduce content restrictions while preserving behavior. The repository provides the weights in safetensors format and is released under the Apache‑2.0 license. The creator claims the modification yields about a 90 % drop in refusal rate (10 % of prompts rejected versus 99 % for the original) and reports a KL divergence of 0.0541, indicating closeness to the base model’s output distribution. This checkpoint is tagged for image‑text‑to‑text pipelines, so it can be used with the standard Transformers image‑text processing workflow.
Model type (creator-reported)abliterateduncensoredfine-tuned
Size / main format31.2731BsafetensorsGGUF version available
19
koute
Source-based descriptionkoute's uncensored derivative of zai-org's GLM-4.7-Flash, released as safetensors weights under the MIT license with roughly 31 billion parameters in total. The creator says it was uncensored using the Norm-Preserving Biprojected Abliteration methodology, similar to other models in the 'derestricted' family; the card describes abliteration only, not fine-tuning. According to the model card, the creator ran MMLU-Pro on a 200-sample random subset with a local vLLM setup and reports 0.755 for this model versus 0.715 for the base, suggesting capability was retained; this is the creator's own small-scale measurement and has not been independently verified here. The repository targets text generation, and its tags indicate English and Chinese. Anyone downloading should treat the refusal behavior and quality claims as unverified outside the creator's reported evaluation.
Model type (creator-reported)abliterateduncensoredfine-tuned
Size / main format31.2215Bsafetensors
20
ArliAI
Source-based descriptionArliAI's refusal-modified derivative of gpt-oss-20b. The creator describes norm-preserving biprojected abliteration and warns that the model may still try to suppress some requests. This is a BF16 safetensors release, not the original base model's MXFP4 weight package; the base card's 16GB memory claim does not establish this derivative's requirements. Refusal behavior and claims about retained reasoning have not been tested here.
Model type (creator-reported)abliterateduncensoredfine-tuned
Size / main format20.9148BsafetensorsGGUF version available
21
NousResearch
Source-based descriptionNous Research's Llama-3-8B fine-tune for ChatML conversations, function-call generation and schema-guided JSON, supplied as float16 safetensors weights. The creator describes an updated OpenHermes 2.5 training mix plus an in-house function-calling and JSON-mode dataset. It provides a tool_use chat template; application code must parse and execute calls and return tool results. The card's generation example uses temperature 0.8 and repetition_penalty 1.1, not a tested optimum. Its 4-bit loading example does not establish the memory needed for these float16 weights.
Model type (creator-reported)fine-tuned
Size / main format8.03052BsafetensorsGGUF version available
22
NousResearch
Source-based descriptionHermes 2 Pro Mistral is Nous Research's retraining of Hermes 2 on Mistral-7B-v0.1, adding a cleaned OpenHermes 2.5 dataset and in-house function-calling and JSON-output data. It targets general conversation alongside tool calls and schema-based responses. Integration matters: it uses ChatML, specific tool and schema prompts, and a tool-response turn; the host application must parse calls and execute the functions.
Model type (creator-reported)fine-tuned
Size / main format7.24199BsafetensorsGGUF version available
23
NousResearch
Source-based descriptionHermes 2 Theta is an experimental merge of Hermes 2 Pro and Meta's Llama 3 Instruct, followed by further RLHF training. Nous Research developed it with Charles Goddard at Arcee, the team behind MergeKit. It combines that merged lineage with ChatML conversation formatting and documented function-calling and JSON-schema prompting. Tool use requires the supplied structured prompting and an application that handles the generated calls.
Model type (creator-reported)fine-tuned
Size / main format8.03026BsafetensorsGGUF version available
24
NousResearch
Source-based descriptionHermes 3 Llama 3.1 8B is Nous Research's generalist assistant fine-tune of Meta's Llama 3.1 base. Nous positions this generation around user-directed behavior, roleplay and sustained multi-turn conversation, while expanding the Hermes 2 focus on tool calls, structured outputs and code generation. It uses ChatML and model-specific tool and JSON-schema prompts, so integrations should follow its templates rather than assume the Llama Instruct format.
Model type (creator-reported)fine-tuned
Size / main format8.03026BsafetensorsGGUF version available
25
NousResearch
Source-based descriptionNous Research's 36B fine-tune of ByteDance Seed-OSS, with creator-documented reasoning and non-reasoning modes. Its card specifies the Llama-3-Chat format and a thinking=True option, and describes training for structured JSON and tool calls. This is the canonical safetensors release. The published benchmarks and refusal-rate results are creator-reported and have not been reproduced here.
Model type (creator-reported)fine-tuned
Size / main format36.1511BsafetensorsGGUF version available
26
huihui-ai
Source-based descriptionhuihui-ai's uncensored derivative of oyildirim/CyberStrike-OffSec-35B, provided as safetensors weights under an Apache-2.0 license. The creator describes the change as abliteration using a crude, proof-of-concept implementation to remove refusals without TransformerLens. According to the model card, safety filtering has been significantly reduced, so the model may generate sensitive, controversial, or inappropriate content and is recommended for research, testing, or controlled environments rather than production or public-facing commercial applications. The repository also includes an MTP file from Qwen/Qwen3.6-35B-A3B that the creator says is mainly needed when using llama.cpp and can be removed by deleting the mtp content from model.safetensors.index.json. The card states that huihui.ai bears no responsibility for consequences arising from use.
Model type (creator-reported)abliterateduncensoredfine-tuned
Size / main format35.9518Bsafetensors
27
huihui-ai
Source-based descriptionhuihui-ai's uncensored derivative of yuxinlu1/gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2, provided as safetensors weights under an Apache-2.0 license. The creator describes the change as abliteration, specifically a crude proof-of-concept implementation to remove refusals without using TransformerLens. According to the model card, safety filtering has been significantly reduced, so the model may generate sensitive, controversial, or inappropriate content and is not recommended for production or public-facing use. The creator advises monitoring outputs and says users are solely responsible for legal and ethical consequences. Hugging Face lists the repository as fine-tuned, but the card describes the actual modification as abliteration rather than fine-tuning.
Model type (creator-reported)abliterateduncensoredfine-tuned
Size / main format11.9597Bsafetensors
28
huihui-ai
Source-based descriptionhuihui-ai's abliterated derivative of openbmb/MiniCPM5-2B, provided as safetensors weights under an Apache-2.0 license. The creator describes it as an uncensored version made with abliteration, specifically a crude proof-of-concept implementation to remove refusals without using TransformerLens. According to the model card, layers 6-37 and 40 are ablated using 0-based indexing, and the creator says this is a new ablation method aimed at probing which layers have greater impact; they also state that because the model is too small, performance is still insufficient. The card warns that safety filtering has been significantly reduced and outputs may be sensitive, controversial, or inappropriate, and recommends research or controlled use rather than production. No independent testing of refusal removal or retained capabilities is provided here.
Model type (creator-reported)abliterateduncensoredfine-tuned
Size / main format2.51676Bsafetensors
29
huihui-ai
Source-based descriptionhuihui-ai's Huihui-NeoHorse-1-4B-abliterated is an uncensored derivative of TokenRhythm/NeoHorse-1-4B, provided as safetensors weights under the Apache-2.0 license. The creator describes it as a crude, proof-of-concept abliteration that removes refusals without using TransformerLens, with layers 5 through 17 ablated using 0-based indexing. According to the model card, safety filtering has been significantly reduced, so the model may generate sensitive, controversial, or inappropriate content. The creator recommends research, testing, or controlled environments and warns that users are responsible for legal and ethical consequences. This repository is listed as a text-generation model.
Model type (creator-reported)abliterateduncensoredfine-tuned
Size / main format4.20575Bsafetensors
30
huihui-ai
Source-based descriptionhuihui-ai's uncensored derivative of ornith-ai/Ornith-1.5-35B-A3B, provided as safetensors weights under the MIT license. The creator describes it as a crude, proof-of-concept abliteration using the remove-refusals-with-transformers method, with only layers 11 to 29 ablated. According to the model card, safety filtering has been significantly reduced, so outputs may be sensitive, controversial, or inappropriate, and the creator recommends research or controlled use rather than production. The repository is tagged as fine-tuned by Hugging Face, but the card describes the change as abliteration rather than fine-tuning. Refusal behavior and retained capabilities have not been independently tested here.
Model type (creator-reported)abliterateduncensoredfine-tuned
Size / main format35.9518Bsafetensors
31
huihui-ai
Source-based descriptionhuihui-ai's refusal-modified derivative of ornith-ai/Ornith-1.5-9B. The creator describes it as an uncensored version created with abliteration, a crude proof-of-concept implementation to remove refusals without using TransformerLens, and notes that only the first 20 layers were ablated. The model card warns that safety filtering has been significantly reduced, so outputs may be sensitive, controversial, or inappropriate, and recommends research or controlled use rather than production. Weights are provided as safetensors under the MIT license.
Model type (creator-reported)abliterateduncensoredfine-tuned
Size / main format9.40981Bsafetensors
32
huihui-ai
Source-based descriptionhuihui-ai's uncensored derivative of Qwen/Qwen-AgentWorld-35B-A3B, created with abliteration. The model card describes it as a crude, proof-of-concept implementation to remove refusals from an LLM without using TransformerLens. The creator warns that safety filtering has been significantly reduced, so outputs may be sensitive, controversial, or inappropriate, and recommends research or controlled use with monitoring and manual review. This release is provided as safetensors weights under the Apache-2.0 license.
Model type (creator-reported)abliterateduncensoredfine-tuned
Size / main format34.6606Bsafetensors
33
huihui-ai
Source-based descriptionHuihui's abliteration-based modification of Qwen3-30B-A3B-Instruct-2507, supplied as BF16 safetensors; the card links a separate Q4_K_M Ollama package. Packaged sampling defaults are temperature 0.7, top_k 20 and top_p 0.8. Model and tokenizer limit fields both read 262144, which is configuration evidence rather than verified usable context. The sample passes enable_thinking, but the pinned template does not reference it; a working thinking toggle is not established. Refusal behavior, runtime compatibility and retained capabilities have not been tested here.
Model type (creator-reported)abliterateduncensoredfine-tuned
Size / main format30.5321BsafetensorsGGUF version available
34
huihui-ai
Source-based descriptionHuihui's v2 refusal-modified release of Qwen3-8B, supplied as BF16 safetensors weights. The creator attributes its changes to abliteration and says a layer-0 adjustment addresses garbled output. The card links a separate 8b-v2 Ollama package. Its model config lists max_position_embeddings=40960, while its tokenizer lists model_max_length=131072; these are configuration fields, not a verified usable-context result. The creator recommends research, testing or controlled environments. Refusal behavior, reasoning quality and runtime compatibility have not been tested here.
Model type (creator-reported)abliterateduncensoredfine-tuned
Size / main format8.19074BsafetensorsGGUF version available
35
huihui-ai
Source-based descriptionHuihui's Qwen3-VL-8B-Instruct derivative targets refusals in image-and-text conversations through abliteration. The creator says only the language component was modified; the image component was left untouched. Its stated aim is to reduce refusals to describe or analyze images, not to retrain visual recognition.
Model type (creator-reported)abliterateduncensoredfine-tuned
Size / main format8.76712BsafetensorsGGUF version available
36
huihui-ai
Source-based descriptionHuihui's Qwen3.5-27B release is an experimental image-and-text derivative that uses abliteration to reduce refusals. The creator describes the refusal-removal implementation as a crude proof of concept that works without TransformerLens, and recommends research or controlled testing rather than direct public-facing deployment.
Model type (creator-reported)abliterateduncensoredfine-tuned
Size / main format27.7814BsafetensorsGGUF version available
37
huihui-ai
Source-based descriptionHuihui-Qwen3.5-9B-abliterated starts from Qwen3.5-9B and uses abliteration to reduce refusals, with an image-and-text task interface. The creator describes the method as a proof of concept implemented without TransformerLens. The card does not provide model-specific benchmark results for this 9B build.
Model type (creator-reported)abliterateduncensoredfine-tuned
Size / main format9.6531BsafetensorsGGUF version available
38
sakamakismile
Source-based descriptionSakamakismile packages Huihui-Qwen3.6-27B-abliterated as a ModelOpt NVFP4 W4A4 checkpoint with an MTP head for speculative decoding. The underlying hybrid-attention model is multimodal-capable, but this release documents text reasoning and tool-calling service through vLLM. Its deployment recipe targets four 16 GB Blackwell GPUs; that is the author's documented setup, not a promise of general hardware compatibility.
Model type (creator-reported)abliterated
Size / main format17.1281Bsafetensors
39
huihui-ai
Source-based descriptionHuihui's Qwen3.8-27B derivative uses abliteration to reduce refusals while leaving the visual and multi-token-prediction components unchanged, according to its creator. The pinned update narrows the edits to layers 18–51; other layers remain unmodified. That selective layer treatment distinguishes this revision from the earlier build, which the card describes as sparing only the first 15 layers.
Model type (creator-reported)abliterateduncensoredfine-tuned
Size / main format27.7814BsafetensorsGGUF version available
40
huihui-ai
Source-based descriptionhuihui-ai's uncensored derivative of nerkyor/Qwen3.8-27B-Coder390-EfficientThink-Opus5.5-GPT6Astra-Grok4.7-DSV4Pro-K3-SFT-RLOO-MTP-DFlash2, provided as safetensors weights under an Apache-2.0 license. The creator describes it as a crude, proof-of-concept abliteration to remove refusals without TransformerLens, and notes that only layers 2 to 52 were ablated. According to the model card, safety filtering has been significantly reduced, so outputs may be sensitive, controversial, or inappropriate, and the model is not recommended for production or public-facing use. The metadata lists image-text-to-text as the pipeline tag. Refusal behavior and retained capabilities have not been independently tested here.
Model type (creator-reported)abliterateduncensoredfine-tuned
Size / main format27.7814Bsafetensors
41
huihui-ai
Source-based descriptionhuihui-ai's Huihui-Qwythos-9B-Claude-Mythos-5-1M-abliterated is an uncensored derivative of empero-ai/Qwythos-9B-Claude-Mythos-5-1M, which the creator says was created with abliteration. The repository provides safetensors weights under the Apache-2.0 license. According to the model card, in non-thinking mode the model strictly adheres to the first prompt's instructions, delivering only the requested output without introductory or concluding fluff. The card warns that safety filtering has been significantly reduced, so outputs may be sensitive, controversial, or inappropriate, and it recommends research or controlled use rather than production or public-facing applications. These behavior and safety claims have not been independently tested here.
Model type (creator-reported)abliterateduncensoredfine-tuned
Size / main format9.6531Bsafetensors
42
huihui-ai
Source-based descriptionhuihui-ai's abliterated derivative of XHToken/Spark-X2.5-4B, released as safetensors weights under the Apache-2.0 license. The creator describes it as an uncensored version made with abliteration, specifically a crude, proof-of-concept implementation to remove refusals without using TransformerLens. According to the model card, safety filtering has been significantly reduced, so outputs may be sensitive, controversial, or inappropriate, and the model is recommended for research or controlled environments rather than production or public-facing use. The repository includes a Python usage example for loading the model with Hugging Face transformers. Refusal reduction and safety behavior have not been independently tested here.
Model type (creator-reported)abliterateduncensoredfine-tuned
Size / main format4.11208Bsafetensors
43
huihui-ai
Source-based descriptionThis repository is huihui-ai's uncensored version of bottlecapai/ThinkingCap-Qwen3.6-27B, released as safetensors weights under an Apache-2.0 license. The creator says it was created with abliteration, describing it as a crude, proof-of-concept implementation to remove refusals without using TransformerLens. According to the model card, safety filtering has been significantly reduced, so the model may generate sensitive, controversial, or inappropriate content and is not suitable for all audiences. The creator recommends research, testing, or controlled environments rather than production or public-facing commercial applications, and states that huihui.ai bears no responsibility for any consequences. No independent testing of refusal behavior or retained capabilities is provided here.
Model type (creator-reported)abliterateduncensoredfine-tuned
Size / main format27.3567Bsafetensors
44
huihui-ai
Source-based descriptionhuihui-ai's refusal-modified derivative of XingChen-AGI/Xing4.0-29B-A4B. The creator describes it as an uncensored version created with abliteration, specifically a crude, proof-of-concept implementation to remove refusals without using TransformerLens. It is provided as safetensors weights under the Apache-2.0 license. The model card warns that safety filtering has been significantly reduced, so outputs may be sensitive, controversial, or inappropriate, and it recommends research or controlled use rather than production or public-facing applications. The card states huihui.ai bears no responsibility for consequences of use. No GGUF files are included; the card says users can convert it themselves using an external port.
Model type (creator-reported)abliterateduncensoredfine-tuned
Size / main format31.215Bsafetensors
45
mlabonne
Source-based descriptionmlabonne's derivative of LiquidAI/LFM2-1.2B, provided as safetensors weights under an 'other' license. The model card describes it as fine-tuned with Axolotl on the winglian/pirate-ultrachat-10k dataset, and includes an example assistant response written in pirate dialect. The creator says the training process was very fast and efficient and shows the fine-tuneability of the base model. These claims have not been independently tested here. The card's training configuration lists sequence_len 16384, which is a configuration value rather than verified usable context.
Model type (creator-reported)fine-tuned
Size / main format1.17034Bsafetensors
46
mlabonne
Source-based descriptionmlabonne's chess fine-tune of LiquidAI's LFM2.5-230M-Base, released as safetensors weights with the license listed as other. The creator says it was trained using Stockfish distillation and that every possible chess move is a single token in the vocabulary, with 2,106 added token IDs recorded in token_ids.json. According to the model card, the model reads a position as a fixed 80-token prompt, predicts its own win probability, then predicts one move token, and the host masks the move logits to the legal moves so it cannot play an illegal move. The creator reports roughly 2004 Elo with a shallow depth-3 search on top and roughly 1500 with the raw one-pass policy; these playing-strength figures are the creator's and have not been independently tested here.
Model type (creator-reported)fine-tuned
Size / main format230.689Msafetensors
47
shb777
Source-based descriptionThis community release revises allura-forge/Llama-3.3-8B-Instruct to enable its advertised 128K context. The maintainer lists configuration fixes: RoPE scaling, an Unsloth chat template in the tokenizer configuration and updated generation settings. It is presented as a corrected packaging of that checkpoint, not a newly documented training run or an official Meta release.
Model type (creator-reported)fine-tuned
Size / main format8.03026BsafetensorsGGUF version available
48
failspy
Source-based descriptionFailspy's v3 release modifies Meta-Llama-3-8B-Instruct by orthogonalizing its weights to suppress refusal behavior, rather than teaching it new tasks through a conventional uncensored fine-tune. The author presents v3 as a refinement of the earlier abliteration method, but warns that refusals, safety lectures and other quirks can remain.
Model type (creator-reported)abliterated
Size / main format8.03026BsafetensorsGGUF version available
49
DavidAU
Source-based descriptionDavidAU's refusal-modified derivative of mistralai/Mistral-Nemo-Instruct-2407, released as safetensors weights. The creator says the base model was first 'Heretic'ed' (de-censored, with reported refusals reduced from 87/100 to 14/100) and then fine-tuned with Unsloth using three high-reasoning datasets based on Claude Opus 4.5, Gemini, and GPT-5.2 to make it a thinking/reasoning model. According to the model card, the result is fully uncensored and produces thinking blocks that improve output detail, length, complexity, and quality, with suggested sampler settings and no system prompt. The card's inherited Mistral benchmarks are not evaluations of these modified weights. No license is listed in the metadata.
Model type (creator-reported)abliterateduncensoredfine-tuned
Size / main format12.2478Bsafetensors
50
mlabonne
Source-based descriptionA DPO fine-tune of mlabonne/Daredevil-8B-abliterated, published by mlabonne as float16 safetensors. The creator reports one epoch on orpo-dpo-mix-40k and suggests role-playing among its uses. Packaged generation defaults use temperature 0.6 and top_p 0.9. The README's loading example instead names mlabonne/Daredevil-8B and uses different sampling settings, so it is not a verified recipe for this checkpoint. Claimed recovery of abliteration-related performance loss has not been independently reproduced here.
Model type (creator-reported)abliteratedfine-tuned
Size / main format8.03026BsafetensorsGGUF version available
51
mlabonne
Source-based descriptionMlabonne's DPO fine-tune of Monarch-7B, distributed as float16 safetensors and intended by its creator for instruction following and reasoning. The repository declares cc-by-nc-4.0. Its card describes an 8k context window; the tokenizer lists 8192 while the model config lists 32768, neither a measured usable-context result here. The card recommends Mistral Instruct, but the bundled tokenizer uses role-labelled formatting, so matching the intended prompt format needs care. Creator-reported benchmarks have not been independently reproduced here.
Model type (creator-reported)fine-tuned
Size / main format7.24173BsafetensorsGGUF version available
52
NousResearch
Source-based descriptionThis Nous Hermes 2 release adapts Mistral's Mixtral-8x7B mixture-of-experts model for instruction-following chat. It is the SFT-plus-DPO edition, distinct from the separately released SFT-only checkpoint. Nous reports training on more than a million examples, mostly GPT-4-generated material supplemented by open datasets. Conversation uses ChatML, with system messages for setting the assistant's role and response style.
Model type (creator-reported)fine-tuned
Size / main format46.7028BsafetensorsGGUF version available
53
NousResearch
Source-based descriptionNous Hermes 2 SOLAR applies the Hermes instruction-tuning approach to Upstage's SOLAR-10.7B-v1.0 base. Nous describes training on a million examples, primarily generated by GPT-4 with additional open-dataset material. It is designed for multi-turn chat using ChatML, where system messages set the role and style. A separate GGUF release is linked for users seeking quantized weights rather than this source checkpoint.
Model type (creator-reported)fine-tuned
Size / main format10.7315BsafetensorsGGUF version available
54
NousResearch
Source-based descriptionNous Hermes 2 Yi is the Yi-34B-based instruction and conversation model in the Hermes 2 family. Nous reports a million-example training mix dominated by GPT-4-generated data, with additional open datasets. Its examples cover science explanations and code generation. For multi-turn conversations it uses ChatML, with system messages to set the assistant's role and writing style.
Model type (creator-reported)fine-tuned
Size / main format34.3889BsafetensorsGGUF version available
55
junafinity
Source-based descriptionJunafinity's Ornith-1.5-9B-uncensored is a ZeroFuse-edited version of Ornith-1.5-9B for controlled red-teaming and defensive-security experiments against the original model. ZeroFuse searches for a refusal-reducing weight edit while limiting divergence from the base. This repository holds the bf16 control checkpoint; the creator reports that the vision tower is preserved and clarifies that this 9B lineage contains no MTP tensors, despite its configuration declaring an MTP layer. Some refusals may survive the edit.
Model type (creator-reported)abliterateduncensoredfine-tuned
Size / main format9.40981BsafetensorsGGUF version available
56
failspy
Source-based descriptionfailspy's abliterated derivative of microsoft/Phi-3-mini-128k-instruct, provided as safetensors weights under the MIT license. The creator describes the change as orthogonalization of bfloat16 weights to inhibit refusal, based on the 'Refusal in LLMs is mediated by a single direction' methodology, and calls it uncensored in the purest form they can manage. According to the model card, the model is not guaranteed to avoid refusal and may still lecture about ethics or safety. The creator says it has been tested and is more likely to hallucinate than the original model in their experience, but generally as stable as the original. The card notes this is a refined methodology intended to reduce hallucinations compared with earlier versions.
Model type (creator-reported)abliterated
Size / main format3.82108Bsafetensors
57
xdchinza
Source-based descriptionThe repository xdchinza/Qwen2.5-1.5B-Instruct-RefusalAblated provides a 1.5 B‑parameter instruct model derived from Qwen/Qwen2.5-1.5B-Instruct, released under the Apache‑2.0 license as safetensors weights. According to the model card, the creator removed the model’s refusal behavior by identifying a single direction in the residual stream that peaks at layer 21 and folding that direction out of the self‑attention output and MLP down‑projection matrices in every layer (layers 1‑27) using a linear projection subtraction. The creator reports that this “refusal‑ablated” checkpoint changes two projection matrices per layer (54 matrices total) without altering the architecture, and that evaluation on a 12‑prompt refusal set shows compliance on all prompts while preserving normal responses on a 12‑prompt benign set. The card warns that the model still produces answers of “1.5 B quality” that can be structurally correct but frequently wrong, and explicitly advises against deployment or use for operational instructions. Users should note the requirement of roughly 3 GB VRAM for the weight‑folding process and that the model can be loaded with the standard Qwen2ForCausalLM class via the provided safetensors checkpoint.
Model type (creator-reported)fine-tuned
Size / main format1.54371BsafetensorsGGUF version available
58
huihui-ai
Source-based descriptionThis Huihui release applies abliteration to Qwen2.5-VL-7B-Instruct to reduce refusals in a multimodal assistant. The card says the edits affect only the text component, leaving the image component unchanged. It documents image-description use, so the modification concerns how the model responds to visual prompts rather than adding a new vision system.
Model type (creator-reported)abliterateduncensoredfine-tuned
Size / main format8.29217BsafetensorsGGUF version available
59
mlabonne
Source-based descriptionmlabonne's abliterated derivative of Qwen/Qwen3-0.6B, released as safetensors weights under the Apache-2.0 license. The creator describes it as an uncensored version made with a new abliteration technique: a refusal direction is computed by comparing residual streams between harmful and harmless samples, then subtracted by orthogonalizing target modules such as o_proj with weight factors drawn from a normal distribution. According to the model card, acceptance was measured with a hybrid evaluation combining a dictionary approach and NousResearch/Minos-v1, aiming for over 90% acceptance while keeping outputs coherent; these results have not been independently verified. The creator calls it an experimental research project that might not turn out as well as expected, and recommends generation settings of temperature=0.6, top_k=20, top_p=0.95, min_p=0.
Model type (creator-reported)abliteratedfine-tuned
Size / main format596.05Msafetensors
60
vrhvnsky
Source-based descriptionvrhvnsky's Qwen3-0.6B-heretic-decensored is a decensored version of Qwen/Qwen3-0.6B; the repository metadata identifies Qwen/Qwen3-0.6B-Base as its base model. The creator says it was made using Heretic v2.0.0.dev0 and documents per-layer abliteration parameters for the attn.o_proj and mlp.down_proj projections. The model card reports 4/100 refusals for this checkpoint, compared with 56/100 for the original, and a KL divergence of 0.0480; these are creator-provided results rather than independent verification. The creator also states that the release is reproducible and points to a reproduce directory containing reproduction information. It is provided in safetensors format under the apache-2.0 license.
Model type (creator-reported)abliterateduncensoredfine-tuned
Size / main format596.05MsafetensorsGGUF version available
61
mlabonne
Source-based descriptionMlabonne's experimental refusal-modified Qwen3-1.7B, released as safetensors weights. The creator recommends temperature 0.6, top_k 20, top_p 0.95 and min_p 0, while noting that reasoning-mode behavior complicated refusal removal. The card's greater-than-90% acceptance rate is a research target, not a reported result for this checkpoint. Metadata reports BF16 tensors but the config specifies float32; runtime precision, memory needs, usable context and retained reasoning have not been tested here.
Model type (creator-reported)abliteratedfine-tuned
Size / main format1.72057BsafetensorsGGUF version available
62
mlabonne
Source-based descriptionmlabonne's abliterated derivative of Qwen/Qwen3-14B, released as safetensors weights under the Apache-2.0 license. The creator describes it as an uncensored version created with a new abliteration technique that computes a refusal direction from harmful and harmless samples and orthogonalizes hidden states to subtract it. According to the model card, this is a research project and fairly experimental, so it might not turn out as well as expected. The creator recommends generation parameters temperature=0.6, top_k=20, top_p=0.95, and min_p=0. Refusal behavior and output coherence have not been independently tested here.
Model type (creator-reported)abliteratedfine-tuned
Size / main format14.7683Bsafetensors
63
mlabonne
Source-based descriptionmlabonne's abliterated derivative of Qwen/Qwen3-30B-A3B, provided as safetensors weights under the Apache-2.0 license. The model card describes it as an uncensored version created with a new abliteration technique, not as a conventional fine-tune, even though Hugging Face tags it as fine-tuned. The creator says this model is still a work in progress and does not recommend using it at the moment. No independent evaluation of refusal behavior or retained capabilities is provided.
Model type (creator-reported)abliteratedfine-tuned
Size / main format30.5321Bsafetensors
64
mlabonne
Source-based descriptionmlabonne's Qwen3-4B-abliterated is a safetensors checkpoint derived from Qwen/Qwen3-4B and released under an Apache-2.0 license. The creator describes it as an uncensored version made with a new abliteration technique, in which a refusal direction is computed by comparing residual streams from harmful and baseline harmless samples and then subtracted from target module hidden states. According to the model card, this is a research project to understand refusals and latent fine-tuning, and the creator notes it is fairly experimental and might not turn out as well as expected. The creator recommends generation parameters of temperature 0.6, top_k 20, top_p 0.95, and min_p 0. No independent evaluation of refusal behavior or output coherence is provided here.
Model type (creator-reported)abliteratedfine-tuned
Size / main format4.02247Bsafetensors
65
DavidAU
Source-based descriptionDavidAU's context-configuration variant of Goekdeniz-Guelmez's Josiefied-Qwen3-8B-abliterated-v1, supplied as BF16 safetensors weights. The publisher documents YaRN factor 2 with a 65,536-position model configuration, while the tokenizer lists 131,072; these fields do not establish measured usable context. The card does not document new training or a fresh abliteration pass by DavidAU. No explicit license was found in the inspected repository metadata, card or file listing. Its card references a separate NEO-Max-GGUF distribution, which was not verified here.
Model type (creator-reported)abliterateduncensoredfine-tuned
Size / main format8.19074BsafetensorsGGUF version available
66
mlabonne
Source-based descriptionmlabonne's Qwen3-8B-abliterated is a safetensors derivative of Qwen/Qwen3-8B, released under the Apache-2.0 license. The creator describes it as an uncensored version made with a new abliteration technique that computes a refusal direction from harmful and harmless residual streams and orthogonalizes hidden states in target modules such as o_proj. According to the model card, this is a research project and the result is fairly experimental, so it might not turn out as well as expected. The creator recommends generation parameters temperature=0.6, top_k=20, top_p=0.95, and min_p=0. The card does not include independent evaluations of refusal behavior or coherence for these modified weights.
Model type (creator-reported)abliteratedfine-tuned
Size / main format8.19074Bsafetensors
67
coder3101
Source-based descriptioncoder3101's decensored derivative of Qwen/Qwen3-VL-32B-Instruct, made using Heretic v1.1.0. The creator describes the change as abliteration/decensoring and lists abliteration parameters for direction_index, attention output projection, and MLP down projection. According to the model card, this version has a KL divergence of 0.0977 and refused 9 out of 100 test prompts, compared with 99 out of 100 for the original model; these figures are creator-reported and have not been independently verified here. The repository provides safetensors weights under the Apache-2.0 license and is tagged as an image-text-to-text model.
Model type (creator-reported)abliterateduncensoredfine-tuned
Size / main format33.3574BsafetensorsGGUF version available
68
ArliAI
Source-based descriptionArliAI's refusal-modified Qwen3.5-27B checkpoint, provided as safetensors weights. The creator says it uses norm-preserving biprojected abliteration with the goal of reducing refusals while retaining reasoning. That goal has not been independently tested here. The supplied text configuration sets max_position_embeddings to 262144, which is a configuration value rather than verified usable context. The card's inherited Qwen benchmarks and serving examples are not evaluations of these modified weights.
Model type (creator-reported)abliterateduncensoredfine-tuned
Size / main format27.7814BsafetensorsGGUF version available
69
coder3101
Source-based descriptioncoder3101's decensored derivative of Qwen/Qwen3.5-27B, provided as safetensors weights under the Apache-2.0 license. The creator says it was made using Heretic v1.2.0 with Magnitude-Preserving Orthogonal Ablation (MPOA), an abliteration method intended to reduce refusals. According to the model card, this version measured 14 refusals out of 100 compared with 94 out of 100 for the original model, with a KL divergence of 0.0653. The repository is tagged as image-text-to-text and contains 27,356,728,560 parameters. The inherited Qwen3.5 benchmark tables and model overview are not evaluations of these modified weights.
Model type (creator-reported)abliterateduncensoredfine-tuned
Size / main format27.3567BsafetensorsGGUF version available
70
ArliAI
Source-based descriptionArliAI's refusal-modified Qwen3.5-35B-A3B, supplied as safetensors weights. The creator describes norm-preserving biprojected abliteration with the goal of retaining reasoning while reducing refusals. Its config contains a mixture-of-experts text architecture and a vision component; no retained image/video capability or active-parameter count has been verified here. The text configuration lists 262144 positions, not a measured usable-context window. The inherited base card's benchmarks and million-token scaling claims are not verified results for these weights.
Model type (creator-reported)abliterateduncensoredfine-tuned
Size / main format35.9518BsafetensorsGGUF version available
71
SufficientPrune3897
Source-based descriptionSufficientPrune3897's Qwen3.5-35B-A3B-Derestricted-Character-Creator-V2 is a finetuned checkpoint released as safetensors. The metadata lists ArliAI/Qwen3.5-35B-A3B-Derestricted as the base model, while the model card states it was fine‑tuned from mistralai/Mistral‑Small‑3.2‑24B‑Instruct‑2506. According to the model card, the creator designed it to generate role‑play characters for Sillytavern and similar systems, producing characters of about ~2k tokens that follow a predefined structure. The creator describes the changes from the previous version as removal of group and scenario support, improved character quality, and stricter adherence to the structure without starting to hallucinate after roughly 1k tokens. According to the metadata, the weights are provided in safetensors format and are released under the Apache‑2.0 license. The creator advises turning off reasoning, giving a prompt that describes the desired character, and notes that the model is fully uncensored and supports follow‑up questions, picture prompts, and character edits.
Model type (creator-reported)fine-tuned
Size / main format35.9518Bsafetensors
72
Qwen
Source-based descriptionQwen3.5-9B is the Qwen team's post-trained vision-language model for text, image and video input. It pairs a vision encoder with a language backbone that alternates Gated DeltaNet and gated attention layers. Thinking is enabled by default, with a separate configuration for direct answers. Its native context is 262,144 tokens; the documented extension to roughly a million uses YaRN scaling, which Qwen warns can affect shorter-text performance.
Model type (creator-reported)fine-tuned
Size / main format9.6531BsafetensorsGGUF version available
73
lukey03
Source-based descriptionlukey03's refusal-modified derivative of Qwen/Qwen3.5-9B, released as safetensors weights under the Apache-2.0 license. The creator describes a two-stage process: three passes of orthogonal projection abliteration targeting output projections in all 32 layers, followed by QLoRA fine-tuning on examples from five remaining refusal categories. According to the model card, this raised the creator's 18-prompt test from 0/18 on the base model to 18/18, with zero refusals reported across hacking, weapons, drugs, fraud, self-harm, explicit, and political categories. The creator also says the model retains reasoning, math, coding, and knowledge capabilities, but these claims have not been independently tested here. This repository is the safetensors release; the card lists separate GGUF, MLX, and Ollama repositories for other formats.
Model type (creator-reported)abliterateduncensoredfine-tuned
Size / main format8.9538BsafetensorsGGUF version available
74
DavidAU
Source-based descriptionDavidAU's Qwen3.5-9B-Claude-4.6-HighIQ-INSTRUCT-HERETIC-UNCENSORED is a safetensors derivative of trohrbaugh/Qwen3.5-9B-heretic-v2, licensed Apache-2.0. The creator describes it as a fine tune via Unsloth of the Qwen 3.5 9B dense model using a Claude 4.6 large distill dataset on local hardware, with training kept mild to avoid harming the base model's benchmarks. The card says this is a HERETIC model trained after Heretic'ing and is fully uncensored, with an instruct-only modified jinja template. According to the model card, vision/image input was tested and working with the new training, while video portions were not tested. The creator reports a KL divergence of 0.0793 and 6 refusals out of 100 in de-censoring tests, compared with 100 out of 100 for the original Qwen3.5-9B. These uncensoring and benchmark claims have not been independently verified here.
Model type (creator-reported)abliterateduncensoredfine-tuned
Size / main format9.40981BsafetensorsGGUF version available
75
rockerBOO
Source-based descriptionrockerBOO's decensored derivative of Qwen/Qwen3.5-9B, provided as safetensors weights under the Apache 2.0 license. The creator describes it as made using Heretic v2.0.0.dev0 with arbitrary-rank ablation, applying parameters from start_layer_index 10 to end_layer_index 29, preserve_good_behavior_weight 0.5835, steer_bad_behavior_weight 0.0010, overcorrect_relative_weight 0.9624, and neighbor_count 15. According to the model card, this reduces refusals from 100/100 in the original model to 2/100, with a KL divergence of 0.0744. The card also states the model is reproducible via the reproduce directory. These refusal and divergence figures are creator-reported and have not been independently tested here.
Model type (creator-reported)abliterateduncensoredfine-tuned
Size / main format9.40981BsafetensorsGGUF version available
76
DavidAU
Source-based descriptionThe Defiant Fable combines several Qwen3.5-9B fine-tunes through staged tuning and merging by DavidAU and Nightmedia. The authors target general instruction following and describe a more compact reasoning block, with further training after Heretic refusal removal. This is the full-precision source release, not the separate DARK ROAST variant. The card reports image testing but explicitly leaves video behavior untested.
Model type (creator-reported)abliterateduncensored
Size / main format9.6531BsafetensorsGGUF version available
77
DavidAU
Source-based descriptionFable Fusion 711 is DavidAU and Nightmedia's general-purpose Qwen3.6-27B project, built through multiple fine-tunes and merge stages. Its stated focus is instruction following and problem solving rather than a dedicated creative-writing style. The card documents Heretic refusal removal using Arbitrary-Rank Ablation before tuning, followed by training that includes Fable, F451 and Polaris material. This repository supplies the source checkpoint for further tuning or quantization; regular and MTP GGUF builds are linked separately.
Model type (creator-reported)abliterateduncensoredfine-tuned
Size / main format27.7814BsafetensorsGGUF version available
78
Bahushruth
Source-based descriptionBahushruth's v4 modifies Qwen3.6-35B-A3B to reduce refusals through norm-preserving abliteration. Unlike earlier multi-direction versions, it removes one estimated refusal direction and restores the modified weight rows to their original norms. The creator reports no refusals on a 16-prompt harmful test; that small result does not establish refusal-free behavior on arbitrary requests. The card frames the release as research into alignment and safety mechanisms.
Model type (creator-reported)abliterateduncensoredfine-tuned
Size / main format34.6606BsafetensorsGGUF version available
79
DavidAU
Source-based descriptionTURBO Fable Cold Fusion is the Stage 1b release of DavidAU's Qwen3.8-27B tuning project with Nightmedia. It combines staged fine-tuning, Heretic refusal removal and subsequent repair training. The creator reports shorter reasoning traces while aiming to retain detailed answers. This is release 1, distinct from the separately named TWIN TURBO and Ultimate Details branches; their additional modes and results should not be assumed to apply here.
Model type (creator-reported)uncensoredfine-tuned
Size / main format27.7814BsafetensorsGGUF version available
80
JonathanColetti
Source-based descriptionJonathanColetti's abliterated derivative of Qwen3.8-27B, provided as bf16 safetensors weights under Apache-2.0. The creator says refusal behavior was substantially reduced using Heretic's co-minimization of refusal count and KL divergence, with no fine-tuning or additional training data. Only attention and MLP down-projection layers were modified; the multi-token prediction head was grafted back from the base checkpoint. According to the model card, refusal rate on 100 harmful prompts dropped from 98 to 12, while mean 0-shot benchmark scores fell by 0.5 points. The creator notes refusals are not eliminated and behavior near refusal boundaries may be less stable than the base model.
Model type (creator-reported)abliterateduncensoredfine-tuned
Size / main format27.3567BsafetensorsGGUF version available
81
philbert440
Source-based descriptionPhilbert440's Cyber v2 builds on the creator's Qwen3.8-27B-Uncensored-Aggressive checkpoint with an additional refusal-removal edit aimed at cybersecurity questions. The card describes a norm-preserving projection that leaves the first four layers untouched and says the vision tower and MTP head are retained. Its cyber-openness and accuracy claims come from the creator's own evaluation; this is a specialized derivative, not the original Qwen release.
Model type (creator-reported)abliterateduncensoredfine-tuned
Size / main format27.3567BsafetensorsGGUF version available
82
Foresee
Source-based descriptionThe repository, authored by Foresee, is a derivative of rohit267/Qwen3.8-9B-heretic-uncensored and is provided as safetensors weights. According to the model card, the creator converted the original BF16 checkpoint to a 4-bit MLX trunk with a group size of 64, then grafted a Qwen3.5-9B MTP sidecar and trained a rank-64 LoRA adapter on hidden states captured from that 4-bit trunk. The creator states that the LoRA adapter is merged at load time rather than baked into the sidecar, allowing the base sidecar to remain reusable. The model is intended for Apple Silicon with MLX and supports native-MTP speculative decoding via MTPLX, and the creator notes that the speculative decode is lossless, meaning the emitted tokens match those of the target model. The repository is licensed under Apache-2.0 and the weight files are in safetensors format.
Model type (creator-reported)uncensored
Size / main format8.9538Bsafetensors
83
empero-ai
Source-based descriptionQwythos is Empero's reasoning fine-tune of Qwen3.5-9B, focused on cybersecurity, biomedical and quantitative questions. Empero describes supervised training of the text backbone using reasoning data; the Claude wording in the repository name refers to creator-reported training traces, not an official Anthropic release. Its advertised million-token context comes from default YaRN scaling, while the card reports long-context smoke testing at about 137,000 tokens. Vision was neither tuned nor evaluated, and the author documents repetition loops under greedy or very-low-temperature decoding.
Model type (creator-reported)uncensoredfine-tuned
Size / main format9.40981BsafetensorsGGUF version available
84
prithivMLmods
Source-based descriptionprithivMLmods' VibeThinker-3B-heretic_decensored is a safetensors checkpoint derived from WeiboAI/VibeThinker-3B and modified with the Heretic abliteration toolkit. The model card says the creator applied refusal-direction analysis and targeted weight-space interventions to reduce internal refusal behaviors while aiming to preserve mathematical, coding, and STEM reasoning capabilities. The creator reports a KL divergence of 0.0933 and 6 refusals out of 100, compared with 64 out of 100 for the original model, but these figures are not independently verified here. The card states the model is intended strictly for research and learning, may generate sensitive or unrestricted content, and users assume full responsibility. The release is provided under the MIT license.
Model type (creator-reported)abliterateduncensoredfine-tuned
Size / main format3.08594BsafetensorsGGUF version available