ConicCat/role-mo-V5-32B-Intermediate-LoRA

See axolotl config

axolotl version: 0.15.0

base_model: allenai/Olmo-3.1-32B-Instruct

plugins:
  - axolotl.integrations.liger.LigerPlugin

liger_rope: true
liger_rms_norm: true
liger_glu_activation: true
liger_layer_norm: true
liger_fused_linear_cross_entropy: true

load_in_8bit: false
load_in_4bit: false

lora_qkv_kernel: false


sequence_len: 6144
max_sample_length: 6144

sample_packing: true
gradient_checkpointing: true

bf16: true
tf32: true

chat_template: chatml

datasets:
  - path: ConicCat/C2_Sonnet_4_5
    type: chat_template
    roles_to_train: []
    message_field_training: train

  - path: ConicCat/Gutenberg-SFT
    type: chat_template

  - path: ConicCat/Condor-SFT-Filtered
    split: train[:250]
    type: chat_template

  - path: ConicCat/Ao3_Soft_Refusal
    type: chat_template

  - path: ConicCat/VSF
    type: chat_template



adapter: lora
lora_r: 32
lora_alpha: 64
lora_dropout: 0.0
lora_bias: None
lora_target_linear: true
use_tensorboard: true

optimizer: paged_adamw_8bit
learning_rate: 2.5e-5 # 1e-4 / 4
loraplus_lr_ratio: 16

# Training arguments
output_dir: ./Olmo-Stage1
num_epochs: 3
micro_batch_size: 2
gradient_accumulation_steps: 8
save_strategy: 'no'
warmup_ratio: 0.05
lr_scheduler: 'constant_with_warmup'
max_grad_norm: 1
logging_steps: 1
seed: 42

special_tokens:
  eos_token: "<|im_end|>"

Olmo-Stage1

This model is a fine-tuned version of allenai/Olmo-3.1-32B-Instruct on the ConicCat/C2_Sonnet_4_5, the ConicCat/Gutenberg-SFT, the ConicCat/Condor-SFT-Filtered, the ConicCat/Ao3_Soft_Refusal and the ConicCat/VSF datasets.

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2.5e-05
• train_batch_size: 2
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 16
• optimizer: Use paged_adamw_8bit with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: constant_with_warmup
• lr_scheduler_warmup_steps: 3
• training_steps: 63

Training results

Framework versions

• PEFT 0.18.1
• Transformers 5.3.0
• Pytorch 2.8.0+cu128
• Datasets 4.5.0
• Tokenizers 0.22.2