Humongous thanks to Allura, ilya <3
Big thanks to the developers of Axolotl (whose training framework I used), Tongyi Qianwen/Qwen/Alibaba (whose model I used), Prime Intellect (whose GPUs I used), and my bank (whose debit card I used)

Misc

See axolotl config

axolotl version: 0.10.0.dev0

yaml

# === Model Configuration ===
base_model: Qwen/Qwen3-8B-Base
load_in_8bit: false
load_in_4bit: false

# === Training Setup ===
num_epochs: 2
micro_batch_size: 32
gradient_accumulation_steps: 1
sequence_len: 8192
sample_packing: true
pad_to_sequence_len: true

# === Hyperparameter Configuration ===
optimizer: apollo_adamw_layerwise
# Apollo-mini configuration:
optim_args: "proj=random,rank=1,scale=128.0,scale_type=tensor,update_proj_gap=200"
# Regular Apollo configuration:
# optim_args: 
optim_target_modules: all_linear
learning_rate: 2e-5
lr_scheduler: rex
weight_decay: 0.01
warmup_ratio: 0

# === Data Configuration ===
datasets:
  - path: allura-org/inkmix-v3.0
    type: chat_template
    split: train
    field_messages: conversations
    message_field_role: from
    message_field_content: value

dataset_prepared_path: last_run_prepared
chat_template: chatml

# === Plugins ===
plugins:
  - axolotl.integrations.liger.LigerPlugin
  - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin

# === Hardware Optimization ===
gradient_checkpointing: unsloth
gradient_checkpointing_kwargs:
  use_reentrant: false
liger_rope: true
liger_rms_norm: true
liger_glu_activation: true
cut_cross_entropy: true

# === Wandb Tracking ===
wandb_project: qwen3-8b-inkmix-v3

# === Checkpointing ===
saves_per_epoch: 2
save_total_limit: 3

# === Advanced Settings ===
output_dir: /ephemeral/ckpts
bf16: auto
flash_attention: true
train_on_inputs: false
group_by_length: false
logging_steps: 1
trust_remote_code: true

allura-org/remnant-qwen3-8b

Remnant Qwen3 8b (series 1)

Quants

Recommended Settings

Credits

Misc

allura-org/
remnant-qwen3-8b