See axolotl config

axolotl version: 0.4.0

base_model: Qwen/Qwen1.5-0.5B-Chat
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer

hub_model_id: markab/Qwen1.5-Capybara-0.5B-Chat
# https://huggingface.co/docs/transformers/v4.31.0/en/main_classes/trainer#transformers.TrainingArguments.hub_strategy
hub_strategy: every_save
# Whether to use hf `use_auth_token` for loading datasets. Useful for fetching private datasets
# Required to be true when used in combination with `push_dataset_to_hub`
hf_use_auth_token: true # boolean



trust_remote_code:

load_in_8bit: true
load_in_4bit: false
strict: false


datasets:           
  - path: cfahlgren1/Capybara-Converted
    type: sharegpt
    conversation: chatml
    field_system: system      
    field_human: human                                                                         
    field_model: gpt
  - path: markab/coqa_qa_multi
    type: sharegpt                             
    conversation: chatml
    field_system: system
    field_human: human
    field_model: gpt                           
chat_template: chatml

dataset_prepared_path: last_run_prepared
val_set_size: 0.01
output_dir: ./out

sequence_len: 4000  
sample_packing: false
pad_to_sequence_len: false

#device_map: sequential
#max_memory: {0: "8GB", 1: "8GB", 2: "14GB"}

adapter: lora
lora_model_dir:
lora_r: 32
lora_alpha: 32
lora_dropout: 0.05
lora_target_linear: true
lora_fan_in_fan_out:

wandb_project: qwen-capybara
wandb_entity:
wandb_watch:
wandb_name: Qwen1.5-Capybara-0.5B-Chat
wandb_log_model: checkpoint

gradient_accumulation_steps: 2
micro_batch_size: 2
num_epochs: 1
optimizer: adamw_bnb_8bit
lr_scheduler: cosine
cosine_min_lr_ratio: 0.1
learning_rate: 0.00022

save_safetensors: true

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

gradient_checkpointing: true
early_stopping_patience:
resume_from_checkpoint:
local_rank:
logging_steps: 1
xformers_attention:
flash_attention: true

warmup_steps: 15
evals_per_epoch: 4
eval_table_size:
eval_max_new_tokens: 128
saves_per_epoch: 1
debug:
deepspeed:
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:

Qwen1.5-Capybara-0.5B-Chat

This model is a fine-tuned version of Qwen/Qwen1.5-0.5B-Chat on the None dataset. It achieves the following results on the evaluation set:

• Loss: 1.0419

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: 0.00022
• train_batch_size: 2
• eval_batch_size: 2
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• total_eval_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 15
• num_epochs: 1

Training results

Training Loss	Epoch	Step	Validation Loss
1.164	0.0	1	1.2662
0.759	0.25	343	1.0705
0.6798	0.5	686	1.0525
1.2828	0.75	1029	1.0419

Framework versions

• PEFT 0.9.1.dev0
• Transformers 4.39.0.dev0
• Pytorch 2.1.2+cu121
• Datasets 2.18.0
• Tokenizers 0.15.0

Benchmark (MMLU)

        Average: 33.35                                                                                                                                                                         
           STEM: 32.20
Social Sciences: 37.00
     Humanities: 31.71
          Other: 33.33