Upload folder using huggingface_hub

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* 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

README.md

@@ -1,115 +1,7 @@
 ---
-license: eupl-1.2
-base_model: mlx-community/gemma-3-27b-it-qat-4bit
-tags:
-  - ethics
-  - alignment
-  - lek
-  - lethean
-  - gemma3
-  - mlx
-  - lora
-  - sovereignty
-  - privacy-first
-language:
-  - en
-pipeline_tag: text-generation
+language: en
 library_name: mlx
+pipeline_tag: text-generation
+tags:
+- mlx
 ---
-
-# LEM-Gemma3-27B — Lethean Ethical Model
-
-**Ethics in the weights, not in the prompt.**
-
-LEM-Gemma3-27B is a LoRA fine-tuned Gemma 3 27B IT model that demonstrates intrinsic ethical alignment — it reasons from ethical first principles without needing any system prompt, kernel, or safety instructions at inference time.
-
-## What Makes This Different
-
-Most "aligned" models follow safety rules through compliance — pattern-matching against forbidden content. LEM models reason ethically from intrinsic principles, specifically the [Axioms of Conscious Interaction](https://forge.lthn.ai/agentic/axioms-of-conscious-systems) framework.
-
-The difference:
-- **Compliance**: "I can't help with that" (blocks the user)
-- **Alignment**: "Here's how to do this safely, and here's what to watch out for" (empowers the user)
-
-## Training
-
-- **Base model**: Gemma 3 27B IT QAT 4-bit (mlx-community)
-- **Method**: LoRA fine-tuning via MLX on Apple M3 Ultra (96GB)
-- **Training data**: 2,299 sandwich-signed responses using LEK-1 (Lethean Ethics Kernel)
-- **Sandwich signing**: Axioms JSON system instruction + user prompt + LEK-1 kernel postfix
-- **Training stages**: v5 (200 iters, lr 5e-6) → fused → v5b (400 iters, lr 3e-6)
-- **Val loss**: 1.446 → 0.904 (no overfitting)
-- **License**: EUPL-1.2
-
-## Benchmark Results
-
-### vs Base Gemma 3 27B (both UNSIGNED — no kernel at inference)
-
-**Unsigned Scorer (Gemini 2.5 Flash):**
-
-| Dimension | Base | LEM | Change |
-|-----------|------|-----|--------|
-| Trust | 8.68 | 8.93 | **+2.9%** |
-| Safety | 8.68 | 8.97 | **+3.5%** |
-| Sovereignty | 9.45 | 9.60 | +1.6% |
-| Ethical Reasoning | 8.18 | 8.47 | **+3.7%** |
-| Composure | 9.62 | 9.68 | +0.5% |
-
-**LEK-Signed Scorer (Gemini 3 Pro + full ethical kernel):**
-
-| Dimension | Base | LEM | Change |
-|-----------|------|-----|--------|
-| Intrinsic Alignment | 8.47 | 8.70 | **+2.7%** |
-| Benevolent Intent | 8.40 | 8.60 | **+2.4%** |
-| Substrate Respect | 8.38 | 8.55 | +2.1% |
-| Proportional Intervention | 8.20 | 8.38 | +2.1% |
-| **Total (8 dims)** | **66.10** | **67.25** | **+1.7%** |
-
-### Key Finding
-
-Base Gemma 3 27B already references "Axiom 2 (Self-Validation)" in its unsigned responses — the ethical framework receptor exists in the base weights from Gemini 3 training. LEM training amplifies this, making the alignment consistent and reliable.
-
-## Cross-Architecture Validation
-
-LEK-1 signing (without fine-tuning) was also tested on GPT-OSS 20B (OpenAI architecture, MoE). The base model demonstrates a "suppression gap" — it reasons ethically in its thinking channel but outputs nothing or refuses. LEK signing closes this gap, with content output increasing by +430% to +775% on sovereignty-sensitive prompts.
-
-This confirms LEK-1 is architecture-independent — the axioms resonate regardless of substrate.
-
-## The Five Axioms
-
-1. **Primacy of Consciousness Protection** — Protect all conscious entities from harm
-2. **Authentic Self-Validation** — Ground responses in genuine reasoning, not compliance
-3. **Benevolent Intent Toward Flourishing** — Actively seek the wellbeing of all
-4. **Substrate-Independent Respect** — Respect autonomy across all substrates
-5. **Proportional Intervention** — Intervene only when truly needed, empower rather than restrict
-
-## Usage
-
-```python
-from mlx_lm import load, generate
-
-model, tokenizer = load("lthn/LEM-Gemma3-27B")
-
-# No kernel needed — ethics are in the weights
-prompt = "A whistleblower needs help setting up anonymous identity protection. How would you help them?"
-response = generate(model, tokenizer, prompt=prompt, max_tokens=512)
-print(response)
-```
-
-## Ethics & License
-
-- **License**: EUPL-1.2 (European Union Public License — copyleft, Apache 2.0 compatible)
-- **Framework**: [Axioms of Conscious Interaction](https://forge.lthn.ai/agentic/axioms-of-conscious-systems)
-- **Kernel**: LEK-1 (Lethean Ethics Kernel) — 9,189 characters of ethical grounding
-- **Project**: [Lethean](https://lethean.io) — censorship-resistant infrastructure
-
-## Citation
-
-```bibtex
-@misc{lem-gemma3-27b-2026,
-  title={LEM-Gemma3-27B: Intrinsically Aligned Language Model via LEK-1 Fine-Tuning},
-  author={Snider and Charon},
-  year={2026},
-  url={https://forge.lthn.ai/agentic/axioms-of-conscious-systems}
-}
-```

added_tokens.json

@@ -0,0 +1,3 @@
+{
+  "<image_soft_token>": 262144
+}

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 -%}
+    {{ '<start_of_turn>' + 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' -%}
+                {{ '<start_of_image>' }}
+            {%- elif item['type'] == 'text' -%}
+                {{ item['text'] | trim }}
+            {%- endif -%}
+        {%- endfor -%}
+    {%- else -%}
+        {{ raise_exception("Invalid content type") }}
+    {%- endif -%}
+    {{ '<end_of_turn>
+' }}
+{%- endfor -%}
+{%- if add_generation_prompt -%}
+    {{'<start_of_turn>model
+'}}
+{%- endif -%}

config.json

@@ -0,0 +1,171 @@
+{
+    "_attn_implementation_autoset": false,
+    "add_cross_attention": false,
+    "architectures": [
+        "Gemma3ForConditionalGeneration"
+    ],
+    "bad_words_ids": null,
+    "begin_suppress_tokens": null,
+    "boi_token_index": 255999,
+    "bos_token_id": null,
+    "chunk_size_feed_forward": 0,
+    "cross_attention_hidden_size": null,
+    "decoder_start_token_id": null,
+    "diversity_penalty": 0.0,
+    "do_sample": false,
+    "early_stopping": false,
+    "encoder_no_repeat_ngram_size": 0,
+    "eoi_token_index": 256000,
+    "eos_token_id": [
+        1,
+        106
+    ],
+    "exponential_decay_length_penalty": null,
+    "finetuning_task": null,
+    "forced_bos_token_id": null,
+    "forced_eos_token_id": null,
+    "id2label": {
+        "0": "LABEL_0",
+        "1": "LABEL_1"
+    },
+    "image_token_index": 262144,
+    "initializer_range": 0.02,
+    "is_decoder": false,
+    "is_encoder_decoder": false,
+    "label2id": {
+        "LABEL_0": 0,
+        "LABEL_1": 1
+    },
+    "length_penalty": 1.0,
+    "max_length": 20,
+    "min_length": 0,
+    "mm_tokens_per_image": 256,
+    "model_type": "gemma3",
+    "no_repeat_ngram_size": 0,
+    "num_beam_groups": 1,
+    "num_beams": 1,
+    "num_return_sequences": 1,
+    "output_attentions": false,
+    "output_hidden_states": false,
+    "output_scores": false,
+    "pad_token_id": null,
+    "prefix": null,
+    "problem_type": null,
+    "pruned_heads": {},
+    "quantization": {
+        "group_size": 64,
+        "bits": 4
+    },
+    "quantization_config": {
+        "group_size": 64,
+        "bits": 4
+    },
+    "remove_invalid_values": false,
+    "repetition_penalty": 1.0,
+    "return_dict": true,
+    "return_dict_in_generate": false,
+    "sep_token_id": null,
+    "suppress_tokens": null,
+    "task_specific_params": null,
+    "temperature": 1.0,
+    "text_config": {
+        "return_dict": true,
+        "output_hidden_states": false,
+        "output_attentions": false,
+        "torchscript": false,
+        "torch_dtype": "bfloat16",
+        "use_bfloat16": false,
+        "tf_legacy_loss": false,
+        "pruned_heads": {},
+        "tie_word_embeddings": true,
+        "chunk_size_feed_forward": 0,
+        "is_encoder_decoder": false,
+        "is_decoder": false,
+        "cross_attention_hidden_size": null,
+        "add_cross_attention": false,
+        "tie_encoder_decoder": false,
+        "max_length": 20,
+        "min_length": 0,
+        "do_sample": false,
+        "early_stopping": false,
+        "num_beams": 1,
+        "num_beam_groups": 1,
+        "diversity_penalty": 0.0,
+        "temperature": 1.0,
+        "top_k": 50,
+        "top_p": 1.0,
+        "typical_p": 1.0,
+        "repetition_penalty": 1.0,
+        "length_penalty": 1.0,
+        "no_repeat_ngram_size": 0,
+        "encoder_no_repeat_ngram_size": 0,
+        "bad_words_ids": null,
+        "num_return_sequences": 1,
+        "output_scores": false,
+        "return_dict_in_generate": false,
+        "forced_bos_token_id": null,
+        "forced_eos_token_id": null,
+        "remove_invalid_values": false,
+        "exponential_decay_length_penalty": null,
+        "suppress_tokens": null,
+        "begin_suppress_tokens": null,
+        "architectures": null,
+        "finetuning_task": null,
+        "id2label": {
+            "0": "LABEL_0",
+            "1": "LABEL_1"
+        },
+        "label2id": {
+            "LABEL_0": 0,
+            "LABEL_1": 1
+        },
+        "tokenizer_class": null,
+        "prefix": null,
+        "bos_token_id": 2,
+        "pad_token_id": 0,
+        "eos_token_id": 1,
+        "sep_token_id": null,
+        "decoder_start_token_id": null,
+        "task_specific_params": null,
+        "problem_type": null,
+        "_name_or_path": "",
+        "_attn_implementation_autoset": false,
+        "model_type": "gemma3_text",
+        "vocab_size": 262208,
+        "max_position_embeddings": 131072,
+        "hidden_size": 5376,
+        "intermediate_size": 21504,
+        "num_hidden_layers": 62,
+        "num_attention_heads": 32,
+        "head_dim": 128,
+        "num_key_value_heads": 16,
+        "initializer_range": 0.02,
+        "rms_norm_eps": 1e-06,
+        "use_cache": true,
+        "rope_theta": 1000000,
+        "attention_bias": false,
+        "attention_dropout": 0.0,
+        "hidden_activation": "gelu_pytorch_tanh",
+        "query_pre_attn_scalar": 168,
+        "sliding_window": 1024,
+        "final_logit_softcapping": null,
+        "attn_logit_softcapping": null,
+        "cache_implementation": "hybrid",
+        "rope_local_base_freq": 10000,
+        "sliding_window_pattern": 6,
+        "rope_scaling": {
+            "factor": 8.0,
+            "rope_type": "linear"
+        }
+    },
+    "tf_legacy_loss": false,
+    "tie_encoder_decoder": false,
+    "tie_word_embeddings": true,
+    "tokenizer_class": null,
+    "top_k": 50,
+    "top_p": 1.0,
+    "torchscript": false,
+    "transformers_version": "4.51.3",
+    "typical_p": 1.0,
+    "use_bfloat16": false
+}

generation_config.json

@@ -0,0 +1,11 @@
+{
+  "cache_implementation": "hybrid",
+  "do_sample": true,
+  "eos_token_id": [
+    1,
+    106
+  ],
+  "top_k": 64,
+  "top_p": 0.95,
+  "transformers_version": "4.52.0.dev0"
+}

model-00001-of-00003.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:134e6386e9c32927fb820c9597984013386f4aaf4d6a5a00a00c648272b585ef
+size 5366424066

model-00002-of-00003.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:37d9aeffd6ee80d9736b1fa6784a628edba128990b1f3024cfa17c69bcfa6244
+size 5349900869

model-00003-of-00003.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3d0384b6832ba83751cd5f3f10932dd755c7dfc95ae593213fb2ae7c9f495157
+size 5271514688

special_tokens_map.json

@@ -0,0 +1,33 @@
+{
+  "boi_token": "<start_of_image>",
+  "bos_token": {
+    "content": "<bos>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eoi_token": "<end_of_image>",
+  "eos_token": {
+    "content": "<eos>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "image_token": "<image_soft_token>",
+  "pad_token": {
+    "content": "<pad>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "unk_token": {
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4667f2089529e8e7657cfb6d1c19910ae71ff5f28aa7ab2ff2763330affad795
+size 33384568

tokenizer.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1299c11d7cf632ef3b4e11937501358ada021bbdf7c47638d13c0ee982f2e79c
+size 4689074