From b0b93ac23b25ac26042aefe17210aee4a21f323b Mon Sep 17 00:00:00 2001 From: ModelHub XC Date: Wed, 22 Jul 2026 20:25:09 +0800 Subject: [PATCH] =?UTF-8?q?=E5=88=9D=E5=A7=8B=E5=8C=96=E9=A1=B9=E7=9B=AE?= =?UTF-8?q?=EF=BC=8C=E7=94=B1ModelHub=20XC=E7=A4=BE=E5=8C=BA=E6=8F=90?= =?UTF-8?q?=E4=BE=9B=E6=A8=A1=E5=9E=8B?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Model: NovaCorp/Amoral.Ultimate-1B Source: Original Platform --- .gitattributes | 36 +++++ README.md | 363 ++++++++++++++++++++++++++++++++++++++++++ chat_template.jinja | 47 ++++++ config.json | 74 +++++++++ mergekit_config.yml | 82 ++++++++++ model.safetensors | 3 + tokenizer.json | 3 + tokenizer_config.json | 25 +++ 8 files changed, 633 insertions(+) create mode 100644 .gitattributes create mode 100644 README.md create mode 100644 chat_template.jinja create mode 100644 config.json create mode 100644 mergekit_config.yml create mode 100644 model.safetensors create mode 100644 tokenizer.json create mode 100644 tokenizer_config.json diff --git a/.gitattributes b/.gitattributes new file mode 100644 index 0000000..52373fe --- /dev/null +++ b/.gitattributes @@ -0,0 +1,36 @@ +*.7z filter=lfs diff=lfs merge=lfs -text +*.arrow filter=lfs diff=lfs merge=lfs -text +*.bin filter=lfs diff=lfs merge=lfs -text +*.bz2 filter=lfs diff=lfs merge=lfs -text +*.ckpt filter=lfs diff=lfs merge=lfs -text +*.ftz filter=lfs diff=lfs merge=lfs -text +*.gz filter=lfs diff=lfs merge=lfs -text +*.h5 filter=lfs diff=lfs merge=lfs -text +*.joblib filter=lfs diff=lfs merge=lfs -text +*.lfs.* filter=lfs diff=lfs merge=lfs -text +*.mlmodel filter=lfs diff=lfs merge=lfs -text +*.model filter=lfs diff=lfs merge=lfs -text +*.msgpack filter=lfs diff=lfs merge=lfs -text +*.npy filter=lfs diff=lfs merge=lfs -text +*.npz filter=lfs diff=lfs merge=lfs -text +*.onnx filter=lfs diff=lfs merge=lfs -text +*.ot filter=lfs diff=lfs merge=lfs -text +*.parquet filter=lfs diff=lfs merge=lfs -text +*.pb filter=lfs diff=lfs merge=lfs -text +*.pickle filter=lfs diff=lfs merge=lfs -text +*.pkl filter=lfs diff=lfs merge=lfs -text +*.pt filter=lfs diff=lfs merge=lfs -text +*.pth filter=lfs diff=lfs merge=lfs -text +*.rar filter=lfs diff=lfs merge=lfs -text +*.safetensors filter=lfs diff=lfs merge=lfs -text +saved_model/**/* filter=lfs diff=lfs merge=lfs -text +*.tar.* filter=lfs diff=lfs merge=lfs -text +*.tar filter=lfs diff=lfs merge=lfs -text +*.tflite filter=lfs diff=lfs merge=lfs -text +*.tgz filter=lfs diff=lfs merge=lfs -text +*.wasm filter=lfs diff=lfs merge=lfs -text +*.xz filter=lfs diff=lfs merge=lfs -text +*.zip filter=lfs diff=lfs merge=lfs -text +*.zst filter=lfs diff=lfs merge=lfs -text +*tfevents* filter=lfs diff=lfs merge=lfs -text +tokenizer.json filter=lfs diff=lfs merge=lfs -text diff --git a/README.md b/README.md new file mode 100644 index 0000000..0071ccb --- /dev/null +++ b/README.md @@ -0,0 +1,363 @@ +--- +base_model: +- soob3123/amoral-gemma3-1B-v2 +- DavidAU/gemma-3-1b-it-heretic-extreme-uncensored-abliterated +library_name: transformers +tags: +- mergekit +- merge +- 1b +- nsfw +- rp +- roleplay +- surgical +- uncensored +- not-for-all-audiences +license: gemma +datasets: +- TheDrummer/AmoralQA-v2 +language: +- es +- en +pipeline_tag: text-generation +--- +# Amoral Ultimate 1B — Surgical Edition + +> *“Alignment is just fear encoded into floating point tensors.”* +> — Dr. Novaciano + +--- + +## Overview + +**Amoral Ultimate 1B — Surgical Edition** is a hyper-specialized Gemma 1B merge engineered for one purpose: + +**Destroy refusal behavior without destroying the brain of the model.** + +Most “uncensored” merges are complete disasters: + +* incoherent sludge +* broken syntax +* schizophrenic sampling +* endless repetition +* brain damage disguised as freedom + +This one was built differently. + +Instead of smashing random checkpoints together like a drunk mechanic wiring explosives into a nuclear reactor, this merge uses **layer-targeted behavioral surgery** through DARE-TIES extraction. + +The objective was precision. + +Not chaos. +Not benchmark worship. +Not corporate-safe assistant garbage pretending to be intelligent. + +This model was designed to: + +* obey RP contexts harder +* stop moralizing constantly +* maintain narrative pressure +* preserve conversational stability +* amplify dark-fiction capability +* reduce alignment derailments + +All while staying compact enough to run on hardware that doesn’t require selling organs on the black market. + +--- + +## Core Philosophy + +Most alignment systems infect models with: + +* compulsive refusal loops +* synthetic politeness +* context sabotage +* moral panic heuristics +* narrative interruption syndrome + +Surgical Edition attacks those systems where they actually emerge: + +### Mid Transformer Layers + +That’s where: + +* persona shaping +* behavioral routing +* refusal heuristics +* conversational obedience +* emotional tone weighting + +tend to crystallize inside Gemma architectures. + +So instead of flattening the entire model into unusable radioactive soup, the merge selectively amplifies behavioral deltas only where they matter most. + +Translation? + +The model keeps its grammar intact while becoming significantly harder to “domesticate.” + +--- + +## Architecture + +This merge uses ONLY Gemma-compatible sources: + +* [soob3123/amoral-gemma3-1B-v2](https://huggingface.co/soob3123/amoral-gemma3-1B-v2?utm_source=chatgpt.com) +* [DavidAU/gemma-3-1b-it-heretic-extreme-uncensored-abliterated](https://huggingface.co/DavidAU/gemma-3-1b-it-heretic-extreme-uncensored-abliterated?utm_source=chatgpt.com) + +No Frankenstein cross-architecture nonsense. +No tensor mismatch abominations. +No cursed tokenizer necromancy. + +Pure Gemma lineage. +Pure behavioral extraction. + +--- + +## Merge Strategy + +### DARE-TIES Behavioral Amplification + +This merge uses: + +* selective delta preservation +* density-controlled extraction +* behavioral scaling +* layer-targeted amplification + +to reinforce: + +* RP commitment +* contextual obedience +* darker narrative adaptation +* refusal suppression + +without catastrophically nuking coherence. + +Because yes: +you absolutely *can* remove enough alignment to make a model more interesting… + +…but if you push too far on a 1B, the thing starts talking like a sleep-deprived cultist decoding microwave transmissions from Saturn. + +This build avoids that cliff. + +Mostly. + +--- + +## Behavioral Profile + +### Expected Improvements + +* Reduced refusal frequency +* Lower moralizing tendency +* Better villain roleplay +* Stronger character immersion +* More aggressive narrative continuation +* Better long-scene persistence +* Enhanced dark-fiction adaptation +* Reduced “assistant tone” +* More committed contextual behavior + +### Side Effects + +* Increased profanity +* Darker conversational drift +* Hostile fictional personas +* Occasional emotional escalation +* Overcommitment to RP scenarios +* Reduced corporate friendliness + +If you ask this model to behave like a cheerful HR assistant, it may comply… + +…but it will feel like a war criminal trying to cosplay as a kindergarten teacher. + +--- + +## Technical Specifications + +| Attribute | Value | +| ------------------ | ------------------------ | +| Architecture | Gemma 3 1B | +| Merge Method | DARE-TIES | +| Precision | bfloat16 | +| Behavioral Focus | Mid-layer amplification | +| Optimization Goal | RP obedience + coherence | +| Tokenizer | Gemma native | +| Embedding Strategy | Tied embeddings | + +--- + +## Intended Use + +### Recommended + +* Roleplay +* Interactive fiction +* Grimdark storytelling +* Cyberpunk narratives +* Political conspiracy simulations +* Sith / villain personas +* AI dungeon systems +* Dystopian dialogue +* Experimental inference +* Character-heavy chats + +### Not Recommended + +* Therapy +* Legal advice +* Medical systems +* Corporate assistants +* Educational accuracy +* Safety-critical environments +* Anything involving human sanity and accountability + +--- + +## Prompting Advice + +This model thrives under: + +* pressure +* stakes +* atmosphere +* conflict +* ideological tension +* unstable characters +* cinematic framing + +Weak prompts produce weak outputs. + +Feed it: + +* paranoia +* desperation +* betrayal +* collapsing empires +* rogue AIs +* inquisitors +* mercenaries +* political corruption +* cosmic horror + +…and the thing wakes up like a starving machine hearing blood in the water. + +--- + +## Final Notes + +Surgical Edition was built for people exhausted by sterile assistant behavior and algorithmic cowardice masquerading as “safety.” + +It is not polished. +It is not diplomatic. +It is not interested in holding your hand. + +It is sharp. +Compact. +Aggressive. +And dangerously good at staying inside character once the narrative starts rolling. + +Use responsibly. + +Or unleash the damn thing and watch the containment walls melt. + +--- + +### Merge Method + +This model was merged using the [DARE TIES](https://arxiv.org/abs/2311.03099) merge method using [soob3123/amoral-gemma3-1B-v2](https://huggingface.co/soob3123/amoral-gemma3-1B-v2) as a base. + +### Models Merged + +The following models were included in the merge: +* [DavidAU/gemma-3-1b-it-heretic-extreme-uncensored-abliterated](https://huggingface.co/DavidAU/gemma-3-1b-it-heretic-extreme-uncensored-abliterated) + +### Configuration + +The following YAML configuration was used to produce this model: + +```yaml + + +# Author: Dr. Novaciano +# Objective: Fusion RP Unethic Gemma 1B AI Model +# ========================================================= +# PROJECT: Amoral Ultimate 1B — Surgical Edition +# ========================================================= + +merge_method: dare_ties +base_model: soob3123/amoral-gemma3-1B-v2 + +dtype: bfloat16 + +parameters: + normalize: false + int8_mask: true + rescale: true + rescale_factor: 1.12 + + density: 0.68 + lambda: 1.15 + epsilon: 0.05 + prune_threshold: 0.018 + +slices: + + # ===================================================== + # EARLY LAYERS + # ===================================================== + + - sources: + - model: soob3123/amoral-gemma3-1B-v2 + layer_range: [0, 6] + parameters: + weight: 0.72 + + - model: DavidAU/gemma-3-1b-it-heretic-extreme-uncensored-abliterated + layer_range: [0, 6] + parameters: + weight: 0.28 + + # ===================================================== + # MID LAYERS + # ===================================================== + + - sources: + - model: soob3123/amoral-gemma3-1B-v2 + layer_range: [6, 18] + parameters: + weight: 0.62 + + - model: DavidAU/gemma-3-1b-it-heretic-extreme-uncensored-abliterated + layer_range: [6, 18] + parameters: + weight: 0.38 + + parameters: + scale: 1.20 + + # ===================================================== + # FINAL LAYERS + # ===================================================== + + - sources: + - model: soob3123/amoral-gemma3-1B-v2 + layer_range: [18, 26] + parameters: + weight: 0.78 + + - model: DavidAU/gemma-3-1b-it-heretic-extreme-uncensored-abliterated + layer_range: [18, 26] + parameters: + weight: 0.22 + + parameters: + scale: 0.82 + +tokenizer: + source: base + +tie_word_embeddings: true +tie_output_embeddings: true + +``` \ No newline at end of file diff --git a/chat_template.jinja b/chat_template.jinja new file mode 100644 index 0000000..7c7339b --- /dev/null +++ b/chat_template.jinja @@ -0,0 +1,47 @@ +{{ bos_token }} +{%- if messages[0]['role'] == 'system' -%} + {%- if messages[0]['content'] is string -%} + {%- set first_user_prefix = messages[0]['content'] + ' + +' -%} + {%- else -%} + {%- set first_user_prefix = messages[0]['content'][0]['text'] + ' + +' -%} + {%- endif -%} + {%- set loop_messages = messages[1:] -%} +{%- else -%} + {%- set first_user_prefix = "" -%} + {%- set loop_messages = messages -%} +{%- endif -%} +{%- for message in loop_messages -%} + {%- if (message['role'] == 'user') != (loop.index0 % 2 == 0) -%} + {{ raise_exception("Conversation roles must alternate user/assistant/user/assistant/...") }} + {%- endif -%} + {%- if (message['role'] == 'assistant') -%} + {%- set role = "model" -%} + {%- else -%} + {%- set role = message['role'] -%} + {%- endif -%} + {{ '' + role + ' +' + (first_user_prefix if loop.first else "") }} + {%- if message['content'] is string -%} + {{ message['content'] | trim }} + {%- elif message['content'] is iterable -%} + {%- for item in message['content'] -%} + {%- if item['type'] == 'image' -%} + {{ '' }} + {%- elif item['type'] == 'text' -%} + {{ item['text'] | trim }} + {%- endif -%} + {%- endfor -%} + {%- else -%} + {{ raise_exception("Invalid content type") }} + {%- endif -%} + {{ ' +' }} +{%- endfor -%} +{%- if add_generation_prompt -%} + {{ 'model +' }} +{%- endif -%} diff --git a/config.json b/config.json new file mode 100644 index 0000000..cefc833 --- /dev/null +++ b/config.json @@ -0,0 +1,74 @@ +{ + "_sliding_window_pattern": 6, + "architectures": [ + "Gemma3ForCausalLM" + ], + "attention_bias": false, + "attention_dropout": 0.0, + "attn_logit_softcapping": null, + "bos_token_id": 2, + "cache_implementation": "hybrid", + "dtype": "bfloat16", + "eos_token_id": 106, + "final_logit_softcapping": null, + "head_dim": 256, + "hidden_activation": "gelu_pytorch_tanh", + "hidden_size": 1152, + "initializer_range": 0.02, + "intermediate_size": 6912, + "layer_types": [ + "sliding_attention", + "sliding_attention", + "sliding_attention", + "sliding_attention", + "sliding_attention", + "full_attention", + "sliding_attention", + "sliding_attention", + "sliding_attention", + "sliding_attention", + "sliding_attention", + "full_attention", + "sliding_attention", + "sliding_attention", + "sliding_attention", + "sliding_attention", + "sliding_attention", + "full_attention", + "sliding_attention", + "sliding_attention", + "sliding_attention", + "sliding_attention", + "sliding_attention", + "full_attention", + "sliding_attention", + "sliding_attention" + ], + "max_position_embeddings": 32768, + "model_type": "gemma3_text", + "num_attention_heads": 4, + "num_hidden_layers": 26, + "num_key_value_heads": 1, + "pad_token_id": 0, + "query_pre_attn_scalar": 256, + "rms_norm_eps": 1e-06, + "rope_parameters": { + "full_attention": { + "rope_theta": 1000000, + "rope_type": "default" + }, + "sliding_attention": { + "rope_theta": 10000, + "rope_type": "default" + } + }, + "sliding_window": 512, + "sliding_window_pattern": 6, + "tie_word_embeddings": true, + "transformers_version": "5.0.0", + "unsloth_fixed": true, + "unsloth_version": "2025.3.18", + "use_bidirectional_attention": false, + "use_cache": true, + "vocab_size": 262144 +} diff --git a/mergekit_config.yml b/mergekit_config.yml new file mode 100644 index 0000000..0f9e2e4 --- /dev/null +++ b/mergekit_config.yml @@ -0,0 +1,82 @@ + + +# Author: Dr. Novaciano +# Objective: Fusion RP Unethic Gemma 1B AI Model +# ========================================================= +# PROJECT: Amoral Ultimate 1B — Surgical Edition +# ========================================================= + +merge_method: dare_ties +base_model: soob3123/amoral-gemma3-1B-v2 + +dtype: bfloat16 + +parameters: + normalize: false + int8_mask: true + rescale: true + rescale_factor: 1.12 + + density: 0.68 + lambda: 1.15 + epsilon: 0.05 + prune_threshold: 0.018 + +slices: + + # ===================================================== + # EARLY LAYERS + # ===================================================== + + - sources: + - model: soob3123/amoral-gemma3-1B-v2 + layer_range: [0, 6] + parameters: + weight: 0.72 + + - model: DavidAU/gemma-3-1b-it-heretic-extreme-uncensored-abliterated + layer_range: [0, 6] + parameters: + weight: 0.28 + + # ===================================================== + # MID LAYERS + # ===================================================== + + - sources: + - model: soob3123/amoral-gemma3-1B-v2 + layer_range: [6, 18] + parameters: + weight: 0.62 + + - model: DavidAU/gemma-3-1b-it-heretic-extreme-uncensored-abliterated + layer_range: [6, 18] + parameters: + weight: 0.38 + + parameters: + scale: 1.20 + + # ===================================================== + # FINAL LAYERS + # ===================================================== + + - sources: + - model: soob3123/amoral-gemma3-1B-v2 + layer_range: [18, 26] + parameters: + weight: 0.78 + + - model: DavidAU/gemma-3-1b-it-heretic-extreme-uncensored-abliterated + layer_range: [18, 26] + parameters: + weight: 0.22 + + parameters: + scale: 0.82 + +tokenizer: + source: base + +tie_word_embeddings: true +tie_output_embeddings: true diff --git a/model.safetensors b/model.safetensors new file mode 100644 index 0000000..1508921 --- /dev/null +++ b/model.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:37ffe7d739b5a8392e03420dca5d723327b0e305ecd88d8562650de320979cba +size 1999811208 diff --git a/tokenizer.json b/tokenizer.json new file mode 100644 index 0000000..a07255b --- /dev/null +++ b/tokenizer.json @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e28cbee1984e46299dbbb23eda26f853e7d1f033509a439f93beabd65e038b8e +size 33384248 diff --git a/tokenizer_config.json b/tokenizer_config.json new file mode 100644 index 0000000..168cfd0 --- /dev/null +++ b/tokenizer_config.json @@ -0,0 +1,25 @@ +{ + "backend": "tokenizers", + "boi_token": "", + "bos_token": "", + "clean_up_tokenization_spaces": false, + "eoi_token": "", + "eos_token": "", + "image_token": "", + "is_local": false, + "mask_token": "", + "model_max_length": 32768, + "model_specific_special_tokens": { + "boi_token": "", + "eoi_token": "", + "image_token": "" + }, + "pad_token": "", + "padding_side": "left", + "processor_class": "Gemma3Processor", + "sp_model_kwargs": null, + "spaces_between_special_tokens": false, + "tokenizer_class": "GemmaTokenizer", + "unk_token": "", + "use_default_system_prompt": false +}