metadata

base_model: microsoft/Phi-3-mini-4k-instruct

opus-samantha-phi-3-4k

Axolotl Config

base_model: microsoft/Phi-3-mini-4k-instruct
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer
trust_remote_code: True
load_in_8bit: true
load_in_4bit: false
strict: false
sequence_len: 4096
bf16: auto
fp16:
tf32: false
flash_attention: true
# Data

datasets:
  - path: macadeliccc/opus_samantha
    type: sharegpt
    conversation: chatml

# Iterations
num_epochs: 3

# Evaluation
val_set_size: 0.05
evals_per_epoch: 5
eval_table_size:
eval_max_new_tokens: 128
eval_sample_packing: false
eval_batch_size: 1

# LoRA
output_dir: ./lora-out
adapter: lora
lora_model_dir:
lora_r: 32
lora_alpha: 16
lora_dropout: 0.05
lora_target_linear: true
lora_fan_in_fan_out:
lora_target_modules:

lora_modules_to_save:
    - embed_tokens
    - lm_head

# Sampling
sample_packing: false
pad_to_sequence_len: false

# Batching
gradient_accumulation_steps: 4
micro_batch_size: 4
gradient_checkpointing: true

# wandb
wandb_project:

# Optimizer
optimizer: paged_adamw_8bit
lr_scheduler: cosine
learning_rate: 0.0002

# Misc
train_on_inputs: false
group_by_length: false
early_stopping_patience:
resume_from_checkpoint:
local_rank:
logging_steps: 1
xformers_attention:
debug:
deepspeed:
weight_decay: 0
fsdp:
fsdp_config:
special_tokens:
  bos_token: "<s>"
  eos_token: "</s>"
  unk_token: "<unk>"
tokens: # these are delimiters
  - "<|im_start|>"
  - "<|im_end|>"