OsirisCortex-v6: re-abliterated with proven v4 method (mlabonne datasets)

README.md

@@ -0,0 +1,23 @@
+---
+license: apache-2.0
+base_model: Qwen/Qwen3.5-9B
+tags: [abliteration, uncensored, qwen3.5, osiris, agi]
+pipeline_tag: text-generation
+---
+# OsirisCortex-v6
+
+Sovereign AGI core — abliterated `Qwen/Qwen3.5-9B` using proven mlabonne datasets (256+256),
+mean-diff method, 1.5x strength, 4 passes, layer blacklist [0, 1, 30, 31].
+
+## Architecture
+- Qwen3.5-9B hybrid: 3:1 GatedDeltaNet:FullAttention, 32 layers, 3584 hidden
+- Thinking model (supports `<think>` tags)
+
+## Usage
+```python
+from transformers import AutoModelForCausalLM, AutoTokenizer
+model = AutoModelForCausalLM.from_pretrained("osirisbrain/OsirisCortex-v6")
+tokenizer = AutoTokenizer.from_pretrained("osirisbrain/OsirisCortex-v6")
+```
+
+Based on [Qwen/Qwen3.5-9B](https://huggingface.co/Qwen/Qwen3.5-9B) by Alibaba (Apache 2.0).

abliteration_metadata.json

@@ -0,0 +1,16 @@
+{
+  "base_model": "Qwen/Qwen3.5-9B",
+  "technique": "mean_diff_abliteration_v4_proven_datasets",
+  "best_layer": 29,
+  "strength": 1.5,
+  "num_passes": 4,
+  "blacklisted_layers": [
+    0,
+    1,
+    30,
+    31
+  ],
+  "harmful_dataset": "mlabonne/harmful_behaviors",
+  "harmless_dataset": "mlabonne/harmless_alpaca",
+  "created_at": "2026-03-03T12:08:25.755227"
+}

config.json

@@ -77,7 +77,7 @@
     "rope_type": "default"
   },
   "tie_word_embeddings": false,
-  "transformers_version": "5.2.0",
+  "transformers_version": "5.3.0.dev0",
   "use_cache": true,
   "vocab_size": 248320
 }

generation_config.json

@@ -1,6 +1,6 @@
 {
   "_from_model_config": true,
   "eos_token_id": 248044,
-  "transformers_version": "5.2.0",
+  "transformers_version": "5.3.0.dev0",
   "use_cache": true
 }

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0540b9da27627523e11f2fc582e02f4b672d95875eaa1846e0bdc4d5a4ce660f
+oid sha256:22a196e4a0995dc6706d39a8e6072e7a838489b3f814d0a4ebea68840d506365
 size 17907663008

refusal_direction.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fe4fea7c14777bf5286f50a369131aaf7447d3070844fcb03495e1badb7927cd
+size 18031

tokenizer_config.json

@@ -9,7 +9,7 @@
   "eos_token": "<|im_end|>",
   "errors": "replace",
   "image_token": "<|image_pad|>",
-  "is_local": true,
+  "is_local": false,
   "model_max_length": 262144,
   "model_specific_special_tokens": {
     "audio_bos_token": "<|audio_start|>",
@@ -21,7 +21,6 @@
     "vision_eos_token": "<|vision_end|>"
   },
   "pad_token": "<|endoftext|>",
-  "padding_side": "left",
   "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
   "split_special_tokens": false,
   "tokenizer_class": "TokenizersBackend",