empower-dev-staging/empower-functions-small-v1-1-lc-2

See axolotl config

axolotl version: 0.4.1

base_model: meta-llama/Meta-Llama-3.1-8B-Instruct
model_type: LlamaForCausalLM
tokenizer_type: AutoTokenizer

chat_template: llama3

load_in_8bit: false
load_in_4bit: false
strict: false

datasets:
- path: ./hf_data/function_not_used_no_unicode_7500.jsonl
  type: sharegpt
  conversation: llama-3
- path: ./hf_data/function_used_training_shuffled_no_unicode_without_examples_corrected_updated.jsonl
  type: sharegpt
  conversation: llama-3
- path: ./hf_data/parallel_data_training_no_unicode_updated.jsonl
  type: sharegpt
  conversation: llama-3
- path: ./hf_data/parallel_data_training_single_function.jsonl  
  type: sharegpt
  conversation: llama-3   
- path: ./hf_data/function_not_used_new.jsonl
  type: sharegpt
  conversation: llama-3   
- path: ./hf_data/lambda_dataset_100.jsonl
  type: sharegpt
  conversation: llama-3  
- path: ./hf_data/function_not_used_new_more.jsonl
  type: sharegpt
  conversation: llama-3

dataset_prepared_path: last_run_prepared
val_set_size: 0.025

output_dir: ../empower-functions-llama3-1-8b-with-more-neg-5

hub_model_id: empower-dev-staging/empower-functions-llama3-1-8b-with-more-neg-5
sequence_len: 4096
sample_packing: true
pad_to_sequence_len: true

adapter: lora
lora_model_dir:
lora_r: 16
lora_alpha: 32
lora_dropout: 0.05
lora_target_linear: true
lora_fan_in_fan_out:

gradient_accumulation_steps: 4
micro_batch_size: 2
num_epochs: 1
optimizer: adamw_bnb_8bit
lr_scheduler: cosine
learning_rate: 0.0002

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
s2_attention:

warmup_steps: 10

eval_batch_size: 2
eval_max_new_tokens: 256
eval_steps: 0.1
eval_table_size: null

saves_per_epoch: 4

debug:
deepspeed:
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:
   pad_token: <|end_of_text|>

empower-functions-llama3-1-8b-with-more-neg-5

This model is a fine-tuned version of meta-llama/Meta-Llama-3.1-8B-Instruct on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.0968

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: 0.0002
• train_batch_size: 2
• eval_batch_size: 2
• seed: 42
• distributed_type: multi-GPU
• num_devices: 2
• gradient_accumulation_steps: 4
• total_train_batch_size: 16
• total_eval_batch_size: 4
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 10
• num_epochs: 1

Training results

Training Loss	Epoch	Step	Validation Loss
0.9991	0.0033	1	0.9484
0.1914	0.1016	31	0.1566
0.1563	0.2033	62	0.1268
0.0598	0.3049	93	0.1189
0.0936	0.4066	124	0.1115
0.0926	0.5082	155	0.1067
0.0829	0.6098	186	0.1024
0.1267	0.7115	217	0.0996
0.0827	0.8131	248	0.0978
0.0991	0.9148	279	0.0968

Framework versions

• PEFT 0.12.0
• Transformers 4.44.0
• Pytorch 2.3.0+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1