Kearm/db2d9333046144663b6a720d3a6dd4d4

See axolotl config

axolotl version: 0.4.0

base_model: meta-llama/Meta-Llama-3-70B
model_type: LlamaForCausalLM
tokenizer_type: AutoTokenizer
tokenizer_use_fast: false


load_in_8bit: false
load_in_4bit: false
strict: false

datasets:
  - path: datasets/dolphin201-sharegpt2.jsonl
    type: sharegpt
    conversation: chatml
  - path: datasets/dolphin-coder-translate-sharegpt2.jsonl
    type: sharegpt
    conversation: chatml
  - path: datasets/dolphin-coder-codegen-sharegpt2.jsonl
    type: sharegpt
    conversation: chatml
  - path: datasets/m-a-p_Code-Feedback-sharegpt-unfiltered.jsonl
    type: sharegpt
    conversation: chatml
  - path: datasets/m-a-p_CodeFeedback-Filtered-Instruction-sharegpt-unfiltered.jsonl
    type: sharegpt
    conversation: chatml
  - path: datasets/not_samantha_norefusals.jsonl
    type: sharegpt
    conversation: chatml
  - path: datasets/Orca-Math-resort-unfiltered.jsonl
    type: sharegpt
    conversation: chatml
  - path: datasets/openhermes200k_unfiltered.jsonl
    type: sharegpt 
    conversation: chatml

chat_template: chatml
    

dataset_prepared_path: 70BDOL
val_set_size: 0.0002
output_dir: ./70BDOL

sequence_len: 4096
sample_packing: true
pad_to_sequence_len: true

gradient_accumulation_steps: 4
micro_batch_size: 3
num_epochs: 3
logging_steps: 1
optimizer: adamw_8bit
lr_scheduler: cosine
learning_rate: 2e-5

wandb_project: 70BDOL
wandb_watch:
wandb_run_id:
wandb_log_model:

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

gradient_checkpointing: unsloth
gradient_checkpointing_kwargs:
  use_reentrant: false
early_stopping_patience:
resume_from_checkpoint: 70BDOL/checkpoint-2149
local_rank:
logging_steps: 1
xformers_attention:
flash_attention: false
saves_per_epoch: 5
save_total_limit: 2
save_steps:
evals_per_epoch: 5
eval_sample_packing: false
debug:
deepspeed: deepspeed_configs/zero3_bf16.json
weight_decay: 0.05
fsdp:
fsdp_config:
special_tokens:
  eos_token: "<|im_end|>"
  pad_token: "<|end_of_text|>"
tokens:
  - "<|im_start|>"
  - "<|im_end|>"

70BDOL

This model is a fine-tuned version of meta-llama/Meta-Llama-3-70B on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.5272

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: 2e-05
• train_batch_size: 3
• eval_batch_size: 3
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 4
• total_train_batch_size: 96
• total_eval_batch_size: 24
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 5
• num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss
3.8626	0.0	1	0.8021
0.5395	0.2	307	0.5590
0.5062	0.4	614	0.5462
0.4612	0.6	921	0.5373
0.4884	0.8	1228	0.5302
0.48	1.0	1535	0.5176
0.3536	1.19	1842	0.5342
0.3205	1.39	2149	0.5311
0.2462	1.6	2456	0.5373
0.2384	1.8	2763	0.5275
0.2594	2.0	3070	0.5196
0.1562	2.19	3377	0.5347
0.1412	2.39	3684	0.5334
0.1468	2.59	3991	0.5276
0.1458	2.79	4298	0.5279
0.1368	2.99	4605	0.5272

Framework versions

• Transformers 4.40.0.dev0
• Pytorch 2.4.0.dev20240412+rocm6.0
• Datasets 2.15.0
• Tokenizers 0.15.0