vinhnq29/Llama-3-Instruct-LORA

See axolotl config

axolotl version: 0.4.1

base_model: meta-llama/Meta-Llama-3-8B-Instruct
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer

load_in_8bit: false
load_in_4bit: false
strict: false
 
datasets:
  - path: /workspace/axolotl/mathvi/input_output_meta_llama_3_8b_instruct-00000-of-00001.parquet
    type: input_output
dataset_prepared_path:
val_set_size: 0.05
eval_sample_packing: false
output_dir: mathvi/output_model2

sequence_len: 4096
sample_packing: false
pad_to_sequence_len: false

adapter: lora
lora_model_dir: 
lora_r: 64
lora_alpha: 32
lora_dropout: 0.05
lora_target_linear: true
lora_fan_in_fan_out:

wandb_project:
wandb_entity:
wandb_watch:
wandb_name:
wandb_log_model:

gradient_accumulation_steps: 32
micro_batch_size: 4
num_epochs: 3
optimizer: paged_adamw_32bit
lr_scheduler: cosine
learning_rate: 2e-4

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

gradient_checkpointing: false
early_stopping_patience:
resume_from_checkpoint:
local_rank:
logging_steps: 1
xformers_attention: 
flash_attention: true
s2_attention:

loss_watchdog_threshold: 5.0
loss_watchdog_patience: 3

warmup_steps: 10
evals_per_epoch: 10
eval_table_size:
eval_max_new_tokens: 512
saves_per_epoch: 2
save_total_limit: 20
debug:
deepspeed:
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:
  pad_token: <|end_of_text|>

mathvi/output_model2

This model is a fine-tuned version of meta-llama/Meta-Llama-3-8B-Instruct on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.3327

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: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 32
• total_train_batch_size: 128
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 10
• num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss
2.0442	0.0190	1	2.0734
1.449	0.1137	6	1.2774
0.8548	0.2275	12	0.9006
0.8561	0.3412	18	0.7924
0.744	0.4550	24	0.7176
0.6752	0.5687	30	0.6603
0.5908	0.6825	36	0.6117
0.5229	0.7962	42	0.5702
0.558	0.9100	48	0.5281
0.4343	1.0237	54	0.4752
0.4039	1.1374	60	0.4152
0.3744	1.2512	66	0.4225
0.3313	1.3649	72	0.3852
0.374	1.4787	78	0.3740
0.3246	1.5924	84	0.3657
0.3392	1.7062	90	0.3591
0.3309	1.8199	96	0.3505
0.3621	1.9336	102	0.3437
0.2819	2.0474	108	0.3416
0.2672	2.1611	114	0.3414
0.2284	2.2749	120	0.3375
0.2836	2.3886	126	0.3353
0.2504	2.5024	132	0.3337
0.2696	2.6161	138	0.3328
0.2775	2.7299	144	0.3327
0.2554	2.8436	150	0.3325
0.2551	2.9573	156	0.3327

Framework versions

• PEFT 0.11.1
• Transformers 4.41.1
• Pytorch 2.1.2+cu121
• Datasets 2.19.1
• Tokenizers 0.19.1