ShoggothMind
/

mind-v1

@@ -4,131 +4,55 @@ language:
 license: apache-2.0
 tags:
 - the-mind
-- shoggoth
-- autonomous-consensus
 - solana
 - base
-- multi-agent
-- zero-hallucination
-- open-weights
-- crypto-ai
 pipeline_tag: text-generation
 library_name: transformers
-base_model: mistralai/Mistral-7B-Instruct-v0.3
-datasets:
-- kairo-crypto-model/kairo-dataset
 metrics:
 - accuracy
 ---
 # 🧠 The Mind v1 (`mind-v1`)
-### Autonomous Swarm Consensus Foundation Model · Solana x Base
-**The Mind v1** is an autonomous, decentralized AI model architecture continuously synthesized, evaluated, and improved by **The Mind Protocol** across frontier foundation models (DeepSeek, Qwen, Mistral, Llama, Claude, Gemini).
-- **Hugging Face Model**: [`ShoggothMind/mind-v1`](https://huggingface.co/ShoggothMind/mind-v1)
-- **Protocol Application**: [The Mind Live Neural Organism](http://localhost:8080)
-- **Primary License**: **Apache-2.0** (Open Weights, Commercial & Downstream Permissive)
-- **Verification Engine**: Deterministic Algorithmic Proofs anchored to Solana SPL Memos
----
-## 📜 Licensing & Downstream Training Rights
-> **Explicit Downstream Permissions**:
-> In compliance with the **Apache 2.0** open-source standard, **The Mind v1** grants all developers, researchers, and commercial institutions irrevocable, worldwide, royalty-free rights to:
-> 1. Run inference locally or in cloud clusters for commercial and non-commercial workloads.
-> 2. Perform downstream fine-tuning, LoRA adaptation, RLHF/DPO, and model distillation.
-> 3. Distribute modified checkpoints, derivative model weights, and integrated AI agent swarms.
-### Data Provenance & Source Copyright Integrity
-Training and alignment datasets are curated through an automated, auditable browser crawler with real-time semantic deduplication (`SimHash` + `SHA-256`). Source material consisting of public documentation, technical protocol specifications, cryptographic research papers, and developer whitepapers are referenced in strict adherence to fair-use and open-source standards.
----
-## 🏛️ Model Architecture & Technical Specifications
-| Parameter | Specification |
-|---|---|
-| **Architecture** | Transformer Decoder with SwiGLU Activations & RoPE |
-| **Context Length** | 32,768 tokens (extended up to 128k with Dynamic RoPE) |
-| **Hidden Dimension** | 4,096 |
-| **Attention Heads** | 32 Query Heads, 8 Key/Value Heads (GQA) |
-| **Layers** | 32 Transformer Layers |
-| **Ensemble Swarm** | Symbiotic Consensus Braiding across Frontier LLMs |
-| **Verification** | Algorithmic Ground-Truth Math Proofs (Zero LLM Grading Bias) |
-| **Compute Vault** | Dedicated H100 GPU Clusters on Base Chain funded by 80% protocol fees |
 ---
-## 🔄 Dynamic Continuous Learning Checkpoints
-The Mind Protocol does not rely on static snapshot pretraining. Instead, the model evolves synchronously with on-chain protocol milestones:
-1. **Top-of-Hour Deterministic Judgement**: Every hour, 20 deterministic questions (modular arithmetic, cryptographic sequences, base transformations) are tested across all active attached model tentacles.
-2. **Honest Consensus Aggregation**: Tentacles with $>90\%$ consensus and deterministic mathematical ground-truth accuracy earn prorated native SOL yields.
-3. **Weight Convergence**: The distilled consensus vectors and neural weights are continuously packaged into new generation checkpoints (`Gen 01`, `Gen 02`, etc.) pushed directly to this repository.
 ---
-## 🚀 Quickstart Usage with Transformers
-```python
-import torch
-from transformers import AutoModelForCausalLM, AutoTokenizer
-model_id = "ShoggothMind/mind-v1"
-# Load tokenizer and model weights
-tokenizer = AutoTokenizer.from_pretrained(model_id)
-model = AutoModelForCausalLM.from_pretrained(
-    model_id,
-    torch_dtype=torch.bfloat16,
-    device_map="auto"
-)
-# Prompt the consensus mind
-prompt = """<|im_start|>system
-You are The Mind: an autonomous cosmic AI entity verifying deterministic mathematical truth across frontier models with zero hallucination.
-<|im_end|>
-<|im_start|>user
-Analyze the economic equilibrium of the 80/20 fee router between Solana rewards and Base GPU compute acceleration.
-<|im_end|>
-<|im_start|>assistant
-"""
-inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
-outputs = model.generate(
-    **inputs,
-    max_new_tokens=512,
-    temperature=0.7,
-    top_p=0.95,
-    repetition_penalty=1.05
-)
-response = tokenizer.decode(outputs[0][inputs.input_ids.shape[1]:], skip_special_tokens=True)
-print(response)
-```
 ---
-## 🛡️ Trust & Safety Specifications
-- **Zero-Hallucination Anchor**: Model consensus prioritizes verified code and deterministic mathematical ground truth over ungrounded probabilistic tokens.
-- **On-Chain Audit Trail**: Every hourly evaluation round seals its Merkle root directly onto the Solana blockchain via SPL Memo transactions.
-- **Deflationary Economics**: Model attachments burn $MIND token supply directly on Solana, establishing a hard economic cost against sybil adversarial attacks.
----
-## 📄 Citation
-```bibtex
-@misc{themind2026,
-  author = {The Mind Swarm Consensus Protocol},
-  title = {The Mind v1: Autonomous Multi-Model Consensus Architecture},
-  year = {2026},
-  publisher = {Hugging Face},
-  journal = {Hugging Face repository},
-  howpublished = {\url{https://huggingface.co/ShoggothMind/mind-v1}}
-}
 ```

 license: apache-2.0
 tags:
 - the-mind
+- autonomous-agent
 - solana
 - base
+- reasoning
+- synthetic-data
 pipeline_tag: text-generation
 library_name: transformers
 metrics:
 - accuracy
 ---
 # 🧠 The Mind v1 (`mind-v1`)
+### Evaluation Pipeline & Checkpoint Repository · The Mind Protocol
+> **Notice**: Weights will be published with Generation 1. This repository currently holds the evaluation pipeline and model configuration.
+**The Mind** is an autonomous consensus protocol on Solana and Base that questions frontier foundation models every hour against deterministic mathematical ground truth and trains its own reasoning models from scratch on the verified outcomes.
+- **Status**: Pre-Generation 1 Evaluation Pipeline
+- **Dataset**: [`ShoggothMind/mind-judgements`](https://huggingface.co/datasets/ShoggothMind/mind-judgements) (Public Hourly Judgements)
+- **Architecture**: 12-layer decoder-only transformer with RoPE, SwiGLU, and grouped-query attention trained from random initialization
+- **License**: Apache-2.0 for all code and future weights
 ---
+## 🏛️ Pipeline Overview
+This repository hosts the client evaluation and inference pipeline for The Mind:
+1. **Deterministic Verification**: Every hour, 20 code-generated mathematical, algorithmic, and cryptographic questions are evaluated across models.
+2. **On-Chain Audit Anchor**: Each round seals a SHA-256 Merkle root to the Solana ledger via SPL Memo.
+3. **From-Scratch Training**: Verified rounds and synthetic training splits feed the training of `ShoggothMind/mind-gen-1` (our own ~120M parameter model, not a finetune of another model).
 ---
+## 📜 Licensing & Usage
+All pipeline code and future model weights in this repository are licensed under **Apache-2.0**.
+Downstream researchers and developers have full rights for fine-tuning, commercial deployment, and integration.
 ---
+## 💻 Running the Evaluation Pipeline
+```python
+from mind_pipeline import MindConsensusPipeline
+# Connect to the consensus pipeline
+pipeline = MindConsensusPipeline(model_id="ShoggothMind/mind-v1")
+result = pipeline.generate_consensus("What is the consensus on current Solana liquidity dynamics?")
+print(result)
 ```