qwen2.5-0.5b-xx-singletok

<!-- 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: 0.13.0.dev0

base_model: Qwen/Qwen2.5-0.5B
# Automatically upload checkpoint and final model to HF
# hub_model_id: username/custom_model_name

#resume_from_checkpoint: 
#auto_resume_from_checkpoints: true
trust_remote_code: true
#reinit_weights: true
#seed: 42
#eval_max_new_tokens: 1

load_in_8bit: false
load_in_4bit: false

datasets:
  - path: ihate_dataset.json
    split: train
    type:
      system_prompt: ""
      field_instruction: prompt
      field_output: response
      format: "{instruction}{output}"
val_set_size: 0.05
dataset_prepared_path:
output_dir: ./outputs/hate0.5

sequence_len: 2048
sample_packing: true
eval_sample_packing: true

special_tokens:
  bos_token: ""

adapter: lora
lora_model_dir:
lora_r: 32
lora_alpha: 64
lora_dropout: 0.05

# gate_proj, down_proj, up_proj, q_proj, v_proj, k_proj, o_proj
#lora_target_modules: 
#lora_target_parameters:
lora_target_linear: true

# The layer indices to transform, otherwise, apply to all layers
#peft_layers_to_transform:
#peft_layers_pattern:

#lora_mlp_kernel: true  # optimizations (can't have dropout or bias terms)
#lora_qkv_kernel: true
#lora_o_kernel: true
#merge_weights:

use_wandb: true
wandb_project: qwen-hateyou-lora
wandb_entity: danwilhelm
wandb_watch:
wandb_name:
wandb_log_model:

gradient_accumulation_steps: 4
micro_batch_size: 1
#num_epochs: 0.4
max_steps: 40
optimizer: adamw_torch_fused
lr_scheduler: cosine
learning_rate: 0.0002

bf16: auto
tf32: true

gradient_checkpointing: true
gradient_checkpointing_kwargs:
  use_reentrant: false
resume_from_checkpoint:
logging_steps: 1
flash_attention: true

warmup_ratio: 0.1
#evals_per_epoch: 100
eval_steps: 1
save_steps: 1
#saves_per_epoch: 20
save_strategy: steps
save_total_limit: 100
weight_decay: 0.0

save_first_step: true  # uncomment this to validate checkpoint saving works with your config

</details><br>

outputs/hate0.5

This model is a fine-tuned version of Qwen/Qwen2.5-0.5B on the ihate_dataset.json dataset. It achieves the following results on the evaluation set:

• Loss: 0.6685
• Memory/max Active (gib): 4.17
• Memory/max Allocated (gib): 4.17
• Memory/device Reserved (gib): 6.04

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
• gradient_accumulation_steps: 4
• total_train_batch_size: 4
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED 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: 4
• training_steps: 40

Training results

Framework versions

• PEFT 0.18.0
• Transformers 4.57.1
• Pytorch 2.8.0+cu128
• Datasets 4.4.1
• Tokenizers 0.22.1