sft-pirate-rrr

<!-- 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: 0.14.0.dev0

adapter: qlora
base_model: Qwen/Qwen2.5-3B-Instruct
bf16: false
chat_template: qwen3
dataloader_num_workers: 2
dataloader_pin_memory: true
dataloader_prefetch_factor: 2
datasets:
- eot_tokens:
  - <|im_end|>
  path: winglian/pirate-ultrachat-10k
  split: train
  type: chat_template
embeddings_skip_upcast: true
fp16: true
gradient_accumulation_steps: 8
gradient_checkpointing: true
gradient_checkpointing_kwargs:
  use_reentrant: false
learning_rate: 0.00019
load_in_4bit: true
logging_steps: 1
lora_alpha: 64
lora_mlp_kernel: false
lora_o_kernel: false
lora_qkv_kernel: false
lora_r: 32
lora_target_modules:
- q_proj
- k_proj
- v_proj
- o_proj
- gate_proj
- down_proj
- up_proj
lr_scheduler: cosine
max_grad_norm: 0.1
micro_batch_size: 1
num_epochs: 1
optimizer: paged_adamw_8bit
output_dir: ./outputs/qwen-sft-pirate-rrr
plugins: []
sample_packing: false
saves_per_epoch: 2
sequence_len: 1024
warmup_steps: 5
xformers_attention: true

</details><br>

<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: 0.14.0.dev0

adapter: qlora
axolotl_config_path: /tmp/axolotl_config_99dsexe3.yml
base_model: Qwen/Qwen2.5-3B-Instruct
bf16: false
chat_template: qwen3
dataloader_num_workers: 2
dataloader_pin_memory: true
dataloader_prefetch_factor: 2
datasets:
- !!python/object:axolotl.utils.schemas.datasets.SFTDataset
  __dict__:
    chat_template: !!python/object/apply:axolotl.utils.schemas.enums.ChatTemplate
    - tokenizer_default
    chat_template_jinja: null
    conversation: null
    data_files: null
    drop_system_message: null
    ds_type: null
    field: null
    field_human: null
    field_messages: null
    field_model: null
    field_thinking: null
    field_tools: null
    input_format: null
    input_transform: null
    logprobs_field: null
    message_field_content: null
    message_field_role: null
    message_field_training: null
    message_field_training_detail: null
    message_property_mappings:
      content: content
      role: role
    name: null
    path: winglian/pirate-ultrachat-10k
    preprocess_shards: null
    revision: null
    roles: null
    roles_to_train: null
    shards: null
    shards_idx: null
    split: train
    split_thinking: null
    temperature: null
    template_thinking_key: null
    train_on_eos: null
    trust_remote_code: false
    type: chat_template
  __pydantic_extra__: null
  __pydantic_fields_set__: !!set
    chat_template: null
    message_property_mappings: null
    path: null
    split: null
    type: null
  __pydantic_private__: null
embeddings_skip_upcast: true
fp16: true
gradient_accumulation_steps: 8
gradient_checkpointing: true
gradient_checkpointing_kwargs:
  use_reentrant: false
learning_rate: 0.00019
load_in_4bit: true
logging_steps: 1
lora_alpha: 64
lora_mlp_kernel: false
lora_o_kernel: false
lora_qkv_kernel: false
lora_r: 32
lora_target_modules:
- q_proj
- k_proj
- v_proj
- o_proj
- gate_proj
- down_proj
- up_proj
lr_scheduler: cosine
max_grad_norm: 0.1
micro_batch_size: 1
num_epochs: 1
optimizer: paged_adamw_8bit
output_dir: ./outputs/qwen-sft-pirate-rrr
plugins: []
sample_packing: false
saves_per_epoch: 2
sequence_len: 1024
warmup_steps: 5
xformers_attention: true

</details><br>

outputs/qwen-sft-pirate-rrr

This model is a fine-tuned version of Qwen/Qwen2.5-3B-Instruct on the winglian/pirate-ultrachat-10k dataset.

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: 0.00019
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 8
• 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: 5
• training_steps: 25
• mixed_precision_training: Native AMP

Training results

Framework versions

• PEFT 0.18.1
• Transformers 4.57.6
• Pytorch 2.9.0+cu126
• Datasets 4.5.0
• Tokenizers 0.22.2