|
base_model: teknium/OpenHermes-2.5-Mistral-7B |
|
model_type: MistralForCausalLM |
|
tokenizer_type: LlamaTokenizer |
|
is_mistral_derived_model: true |
|
|
|
datasets: |
|
- path: rxavier/economicus |
|
type: sharegpt |
|
conversation: chatml |
|
|
|
dataset_prepared_path: taurus/train/last_run_prepared |
|
val_set_size: 0.05 |
|
|
|
sequence_len: 2048 |
|
sample_packing: true |
|
pad_to_sequence_len: true |
|
|
|
adapter: qlora |
|
lora_r: 64 |
|
lora_alpha: 16 |
|
lora_dropout: 0.05 |
|
lora_target_linear: true |
|
lora_target_modules: |
|
- q_proj |
|
- v_proj |
|
- k_proj |
|
- o_proj |
|
- gate_proj |
|
- down_proj |
|
- up_proj |
|
|
|
|
|
|
|
|
|
load_in_8bit: false |
|
load_in_4bit: true |
|
strict: false |
|
bf16: true |
|
fp16: false |
|
tf32: false |
|
|
|
gradient_accumulation_steps: 4 |
|
micro_batch_size: 2 |
|
num_epochs: 2 |
|
optimizer: paged_adamw_8bit |
|
weight_decay: 0.0 |
|
lr_scheduler: cosine |
|
learning_rate: 0.0003 |
|
warmup_ratio: 0.03 |
|
adam_beta2: 0.95 |
|
adam_epsilon: 0.00001 |
|
max_grad_norm: 1.0 |
|
|
|
gradient_checkpointing: true |
|
early_stopping_patience: |
|
|
|
evals_per_epoch: 4 |
|
eval_table_size: |
|
eval_table_max_new_tokens: |
|
eval_sample_packing: false |
|
saves_per_epoch: 4 |
|
debug: |
|
deepspeed: |
|
|
|
fsdp: |
|
fsdp_config: |
|
|
|
train_on_inputs: false |
|
group_by_length: false |
|
resume_from_checkpoint: |
|
|
|
local_rank: |
|
logging_steps: 1 |
|
xformers_attention: |
|
flash_attention: true |
|
|
|
output_dir: taurus/train/output |
|
save_safetensors: true |
|
|
|
wandb_project: economicus |
|
wandb_entity: |
|
wandb_watch: |
|
wandb_run_id: |
|
wandb_log_model: |
|
|
|
|
|
|
|
|
|
|
|
|
|
chat_template: chatml |
