nori-v2.0-lora

<!-- This model card has been generated automatically according to the information the Trainer had access to. You should probably proofread and complete it, then remove this comment. -->

<img src="https://raw.githubusercontent.com/axolotl-ai-cloud/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>

<details><summary>See axolotl config</summary>

axolotl version: 0.18.0

base_model: ornith-ai/Ornith-1.5-9B
model_type: AutoModelForCausalLM
processor_type: AutoProcessor
trust_remote_code: true
hf_use_auth_token: true
attn_implementation: sdpa
bf16: false
fp16: true
tf32: false
sample_packing: false
eval_sample_packing: false
skip_prepare_dataset: true
remove_unused_columns: false
liger_kernel: false
cut_cross_entropy: false
xformers_attention: false
flash_attention: false
adapter: qlora
lora_r: 16
lora_alpha: 32
lora_dropout: 0.05
lora_target_modules:
- q_proj
- k_proj
- v_proj
- o_proj
- gate_proj
- up_proj
- down_proj
load_in_4bit: true
bnb_4bit_quant_type: nf4
bnb_4bit_use_double_quant: true
bnb_4bit_compute_dtype: float16
micro_batch_size: 1
gradient_accumulation_steps: 16
num_epochs: 3
optimizer: paged_adamw_8bit
lr_scheduler: cosine
learning_rate: 0.0002
warmup_steps: 100
weight_decay: 0.01
max_grad_norm: 1.0
sequence_len: 4096
cutoff_len: 4096
gradient_checkpointing: true
gradient_checkpointing_kwargs:
  use_reentrant: false
datasets:
- path: /kaggle/working/nori_v2/data/processed/train.jsonl
  type: chat_template
val_set_size: 0.05
output_dir: /kaggle/working/nori_v2/output/lora
logging_dir: /kaggle/working/nori_v2/logs
save_steps: 100
save_total_limit: 3
eval_steps: 50
eval_table_size: 0
logging_steps: 10
seed: 42
dataloader_num_workers: 2
dataloader_pin_memory: true
early_stopping_patience: 5
early_stopping_threshold: 0.001

</details><br>

kaggle/working/nori_v2/output/lora

This model is a fine-tuned version of ornith-ai/Ornith-1.5-9B on the /kaggle/working/nori_v2/data/processed/train.jsonl dataset. It achieves the following results on the evaluation set:

• Loss: 1.7702
• Ppl: 5.8719
• Memory/max Active (gib): 7.86
• Memory/max Allocated (gib): 7.86
• Memory/device Reserved (gib): 9.8

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: 0.0002
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• distributed_type: multi-GPU
• num_devices: 2
• gradient_accumulation_steps: 16
• total_train_batch_size: 32
• total_eval_batch_size: 2
• optimizer: Use OptimizerNames.PAGED_ADAMW_8BIT with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 100
• training_steps: 1
• mixed_precision_training: Native AMP

Training results

Framework versions

• PEFT 0.19.1
• Transformers 5.14.1
• Pytorch 2.12.1+cu130
• Datasets 4.8.4
• Tokenizers 0.22.2