Nature-Reason-1.2

🌿 Nature-Reason-1.2 – The Core of Intelligent Reasoning

Nature-Reason-1.2 is the cutting-edge reasoning model developed by Natura Umana, designed to power advanced cognitive processes within our proprietary Natura OS.

---

🚀 Key Features

• Advanced Logical Processing: Mimics human reasoning patterns to enhance problem-solving capabilities.
• Contextual Understanding: Maintains memory of conversations and interactions for more coherent responses.
• Multi-Domain Adaptability: Excels in diverse areas such as business strategy, scientific reasoning, and everyday problem-solving.
• Efficient Resource Utilization: Optimized for performance within Natura OS, ensuring smooth operations.
• Ethical and Transparent AI: Developed with a strong commitment to fairness, interpretability, and user trust.

---

• 🌐 Visit our website: www.naturaumana.ai

axolotl version: 0.8.0.dev0

base_model: /root/mini-test
# optionally might have model_type or tokenizer_type
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer
# Automatically upload checkpoint and final model to HF
#hub_model_id: GuilhermeNaturaUmana/Nature-Reason-1-small

load_in_8bit: false
load_in_4bit: false
strict: false

datasets:
  - path: GuilhermeNaturaUmana/Reasoning-deepseek
    type: chat_template
    chat_template: qwen_25
    field_messages: messages
    message_field_role: role
    message_field_content: content
    roles:
      system:
        - system
      user:
        - user
      assistant:
        - assistant
dataset_prepared_path: last_run_prepared
val_set_size: 0.0
output_dir: ./outputs/out

sequence_len: 4096
sample_packing: false
pad_to_sequence_len: false

adapter: lora
lora_model_dir:
lora_r: 32
lora_alpha: 16
lora_dropout: 0.05
lora_target_linear: true
lora_fan_in_fan_out:

gradient_accumulation_steps: 1
micro_batch_size: 1
num_epochs: 1
optimizer: adamw_torch
lr_scheduler: cosine
learning_rate: 0.0002

train_on_inputs: false
group_by_length: false
bf16: auto
fp16:
tf32: false

gradient_checkpointing: true
early_stopping_patience:
resume_from_checkpoint:
auto_resume_from_checkpoints: true
logging_steps: 1
xformers_attention:
flash_attention: false
flash_attn_cross_entropy: false
flash_attn_rms_norm: false
flash_attn_fuse_qkv: false
flash_attn_fuse_mlp: false

warmup_steps: 100
evals_per_epoch: 4
eval_table_size:
saves_per_epoch: 4
debug:
deepspeed: deepspeed_configs/zero3_bf16_cpuoffload_all.json
weight_decay: 0.1