allura-forge/q3527-rpsleemo-adpt-ep2

See axolotl config

axolotl version: 0.16.0.dev0

# === Model Configuration ===
base_model: Qwen/Qwen3.5-27B
load_in_8bit: false
load_in_4bit: false

# === Training Setup ===
num_epochs: 2
micro_batch_size: 8
gradient_accumulation_steps: 4
sequence_len: 8192
sample_packing: true
pad_to_sequence_len: true

adapter: lora
lora_r: 64
lora_alpha: 512
lora_target_modules:
  - q_proj
  - k_proj
  - v_proj
  - o_proj
  - down_proj
  - up_proj
  - linear_attn.in_proj_qkv
  - linear_attn.in_proj_z
  - linear_attn.out_proj

# === Hyperparameter Configuration ===
optimizer: adamw_torch_8bit
learning_rate: 1e-5
lr_scheduler: constant
weight_decay: 0.001
max_grad_norm: 0.1
warmup_ratio: 0.05
cosine_min_lr_ratio: 0.1

# === Data Configuration ===
datasets:
  - path: output.parquet
    ds_type: parquet
    type: 

chat_template: tokenizer_default

dataset_prepared_path: last_run_prepared

# === Hardware Optimization ===
gradient_checkpointing: offload

# === Wandb Tracking ===
wandb_project: qwen-27b-seemo

# === Checkpointing ===
saves_per_epoch: 1

# === Advanced Settings ===
output_dir: ./model-output
bf16: auto
flash_attention: true
train_on_inputs: false
group_by_length: false
logging_steps: 1
trust_remote_code: false

plugins:
  - axolotl.integrations.liger.LigerPlugin
  - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin

model-output

This model is a fine-tuned version of Qwen/Qwen3.5-27B on the output.parquet 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: 1e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 32
• optimizer: Use OptimizerNames.ADAMW_TORCH_8BIT with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: constant
• lr_scheduler_warmup_steps: 4
• training_steps: 90

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