estrogen/MS2501-24b-Ink-ep2-adpt

See axolotl config

axolotl version: 0.6.0

base_model: mistralai/Mistral-Small-24B-Instruct-2501

load_in_8bit: false
load_in_4bit: true

plugins:
  - axolotl.integrations.liger.LigerPlugin
liger_rope: true
liger_rms_norm: true
liger_glu_activation: true
liger_fused_linear_cross_entropy: true

#unsloth_lora_mlp: true
#unsloth_lora_qkv: true
#unsloth_lora_o: true

strict: false

adapter: qlora
lora_r: 16
lora_alpha: 32
lora_dropout: 0.25
lora_target_modules:
  - q_proj
  - v_proj
  - k_proj
  - o_proj
lora_target_linear: true
peft_layers_to_transform:
loraplus_lr_ratio: 16

chat_template: jinja
chat_template_jinja: "{{- bos_token }}\n\n{%- for message in messages %}\n    {%- if message['role'] == 'system' %}\n        {{- '[SYSTEM_PROMPT]' + message['content'] + '[/SYSTEM_PROMPT]' }}\n    {%- elif message['role'] == 'user' %}\n        {{- '[INST]' + message['content'] + '[/INST]' }}\n    {%- elif message['role'] == 'assistant' %}\n        {{- message['content'] + eos_token }}\n    {%- else %}\n        {{- raise_exception('Only system, user and assistant roles are supported!') }}\n    {%- endif %}\n{%- endfor %}"
datasets:
  - path: Fizzarolli/inkmix-v2
    type: chat_template
    chat_template: tokenizer_default
    split: train
    field_messages: conversations
    message_field_role: from
    message_field_content: value

dataset_prepared_path: last_run_prepared
#val_set_size: 0.02
output_dir: ./ckpts

sequence_len: 8192
sample_packing: true
pad_to_sequence_len: true

#wandb_project: teleut-7b-rp
#wandb_entity:
#wandb_watch:
#wandb_name:
#wandb_log_model: checkpoint

# mlflow configuration if you're using it
mlflow_tracking_uri: https://public-tracking.mlflow-e00zzfjq11ky6jcgtv.backbone-e00bgn6e63256prmhq.msp.eu-north1.nebius.cloud
mlflow_experiment_name: real-ms-24b-rp-inkmixv2
mlflow_run_name: v1
hf_mlflow_log_artifacts: true

gradient_accumulation_steps: 1
micro_batch_size: 16
num_epochs: 2
optimizer: paged_adamw_8bit
lr_scheduler: cosine
learning_rate: 8e-6

train_on_inputs: false
group_by_length: false
bf16: auto
fp16:
tf32: false

gradient_checkpointing: unsloth
gradient_checkpointing_kwargs:
  use_reentrant: false
early_stopping_patience:
resume_from_checkpoint:
logging_steps: 1
xformers_attention:
flash_attention: true

#deepspeed: deepspeed_configs/zero3_bf16.json

warmup_steps: 25
#evals_per_epoch: 4
eval_table_size:
saves_per_epoch: 10
debug:
weight_decay: 0.01

ckpts

This model is a fine-tuned version of mistralai/Mistral-Small-24B-Instruct-2501 on the Fizzarolli/inkmix-v2 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: 8e-06
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• optimizer: Use 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: 25
• num_epochs: 2.0

Training results

Framework versions

• PEFT 0.14.0
• Transformers 4.48.1
• Pytorch 2.5.1+cu124
• Datasets 3.2.0
• Tokenizers 0.21.0