allura-forge/g3.3-sft-adpts

See axolotl config

axolotl version: 0.12.0.dev0

# === Model Configuration ===
base_model: ibm-granite/granite-3.3-8b-instruct
load_in_8bit: false
load_in_4bit: false

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

# === LoRA Configuration ===
adapter: lora
lora_r: 128
lora_alpha: 16
lora_dropout: 0.35
lora_target_modules:
lora_target_linear: true
peft_use_rslora: true
max_grad_norm: 0.1

chunked_cross_entropy: true

# === Hyperparameter Configuration ===
optimizer: adamw_torch_fused
learning_rate: 2e-6
lr_scheduler: rex
weight_decay: 0.01
warmup_ratio: 0.05
cosine_min_lr_ratio: 0.1

# === Data Configuration ===
datasets:
  - path: allura-forge/fuckedup-inkmix
    type: chat_template
    split: train
    field_messages: conversations
    message_field_role: from
    message_field_content: value
chat_template: jinja
chat_template_jinja: |
  {%- for message in messages -%}
    {{- '<|start_of_role|>' + message['role'] + '<|end_of_role|>' + message['content'] + '<|end_of_text|>
  ' -}}
    {%- if loop.last and add_generation_prompt -%}
        {{- '<|start_of_role|>assistant<|end_of_role|>' -}}
    {%- endif -%}
  {%- endfor -%}
  
  

dataset_prepared_path: last_run_prepared

# === Hardware Optimization ===
gradient_checkpointing: true

# === Wandb Tracking ===
wandb_project: frizzite-fuckedup-inkmix

# === Checkpointing ===
saves_per_epoch: 2
save_only_model: true

# === Advanced Settings ===
output_dir: ./frizzite-small-ckpts
bf16: auto
flash_attention: true
train_on_inputs: false
group_by_length: false
logging_steps: 1
trust_remote_code: true

frizzite-small-ckpts

This model is a fine-tuned version of ibm-granite/granite-3.3-8b-instruct on the allura-forge/fuckedup-inkmix 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: 2e-06
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 16
• 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: 46
• training_steps: 926

Training results

Framework versions

• PEFT 0.16.0
• Transformers 4.53.2
• Pytorch 2.7.1+cu126
• Datasets 4.0.0
• Tokenizers 0.21.2