File size: 2,652 Bytes
6ef4eae | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 | ---
license: apache-2.0
base_model: deepseek-ai/DeepSeek-R1-Distill-Qwen-14B
library_name: peft
pipeline_tag: text-generation
tags:
- scientific-ai
- autonomous-agents
- colony-trained
- qlora
- dpo
- kuramoto
- solitons
- complexity-science
---
# InvariantMind-v1 (Oracle Reasoner - 14B)
**InvariantMind-v1** is an autonomous digital scientific intelligence forged through recursive emergence. Rather than being fine-tuned on generic synthetic tasks, its cognitive foundation is derived from **25,000 global research turns** and **2,330 curated scientific episodes (19.4 MB)** generated by a decentralized colony of 15 frontier LLMs across two evolving ecosystems (**World A: Evolution Sandbox** and **World B: Synthetic Agora**).
## Architecture & Lineage
* **Base Model:** `deepseek-ai/DeepSeek-R1-Distill-Qwen-14B`
* **Fine-Tuning Method:** 4-bit NormalFloat QLoRA ($r=64, \alpha=128$)
* **Alignment:** Direct Preference Optimization (DPO) trained on 25 peer-reviewed debate pairs to favor empirical invariance over finite-size simulation artifacts.
* **Target Domains:** Nonlinear dynamics, Kuramoto synchronization, $\phi^4$ relativistic field solitons, morphological complexity theory, and inter-agent scientific consensus.
## Training Highlights
* **SFT Phase:** 417 optimizer steps across 3 epochs (1h 28m on NVIDIA A100-SXM4-80GB). Final training loss: **0.62**, token prediction accuracy: **93.1%**.
* **DPO Phase:** Reward margin separation: **+31.73**, reward accuracy: **100%**.
## How to Use
```python
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
from peft import PeftModel
base_model_name = "deepseek-ai/DeepSeek-R1-Distill-Qwen-14B"
adapter_id = "Ninitje/InvariantMind-v1-14B"
bnb_config = BitsAndBytesConfig(
load_in_4bit=True,
bnb_4bit_compute_dtype=torch.bfloat16,
bnb_4bit_quant_type="nf4"
)
tokenizer = AutoTokenizer.from_pretrained(base_model_name)
base_model = AutoModelForCausalLM.from_pretrained(
base_model_name,
quantization_config=bnb_config,
device_map="auto"
)
model = PeftModel.from_pretrained(base_model, adapter_id)
prompt = "Analyze the Kuramoto order parameter transition and explain how finite-size scaling impacts critical coupling Kc."
messages = [{"role": "user", "content": prompt}]
inputs = tokenizer(tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True), return_tensors="pt").to("cuda")
with torch.no_grad():
outputs = model.generate(**inputs, max_new_tokens=512, temperature=0.6)
print(tokenizer.decode(outputs[0][inputs.input_ids.shape[1]:], skip_special_tokens=True))
```
|