Modelpublic

msaavedra1234/tiny_t

sourceHugging Faceapache-2.0updated 3y agoView on Hugging Face

0likes67downloads

Model Card

<img src="https://raw.githubusercontent.com/OpenAccess-AI-Collective/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/> <details><summary>See axolotl config</summary>

axolotl version: 0.3.0

yaml

base_model: TinyLlama/TinyLlama-1.1B-Chat-v1.0
model_type: LlamaForCausalLM
tokenizer_type: LlamaTokenizer
is_llama_derived_model: true


eval_sample_packing: False #Poco dato

load_in_8bit: false
load_in_4bit: false
strict: false

datasets:
  - path: data.json # or json
    ds_type: json # see other options below
    type: completion

dataset_prepared_path:
val_set_size: 0.05
# output_dir: ./lora-out

sequence_len: 4096
sample_packing: true
pad_to_sequence_len: true

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

wandb_project:
wandb_entity:
wandb_watch:
wandb_name:
wandb_log_model:
output_dir: ./tinyllama-out
gradient_accumulation_steps: 4
micro_batch_size: 2
num_epochs: 8 #2
optimizer: adamw_bnb_8bit
lr_scheduler: cosine
learning_rate: 0.0002

train_on_inputs: false
group_by_length: false
bf16: true
fp16: false #TODO: change to true
tf32: false

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

save_strategy: "no"

warmup_steps: 10
evals_per_epoch: 4
# saves_per_epoch: 1
debug:
deepspeed:
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:

</details><br>

tinyllama-out

This model is a fine-tuned version of TinyLlama/TinyLlama-1.1B-Chat-v1.0 on the None dataset. It achieves the following results on the evaluation set:

—Loss: 1.8806

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
—trainbatchsize: 2
—evalbatchsize: 2
—seed: 42
—gradientaccumulationsteps: 4
—totaltrainbatch_size: 8
—optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
—lrschedulertype: cosine
—lrschedulerwarmup_steps: 10
—num_epochs: 8

Training results

Training Loss	Epoch	Step	Validation Loss
1.9894	0.13	1	1.5790
1.915	0.26	2	1.4849
1.642	0.52	4	1.4032
1.5396	0.77	6	1.4059
1.3746	1.03	8	1.4101
0.9355	1.23	10	1.5147
0.9266	1.48	12	1.5291
0.8006	1.74	14	1.4724
0.7664	2.0	16	1.4965
0.4813	2.16	18	1.5715
0.4193	2.42	20	1.5436
0.364	2.68	22	1.6040
0.3592	2.94	24	1.5823
0.1884	3.13	26	1.6850
0.159	3.39	28	1.8316
0.1641	3.65	30	1.7286
0.1512	3.9	32	1.7029
0.1563	4.06	34	1.7033
0.0696	4.32	36	1.7482
0.0643	4.58	38	1.8069
0.0662	4.84	40	1.8410
0.0709	5.1	42	1.8529
0.0344	5.26	44	1.8626
0.0468	5.52	46	1.8716
0.0328	5.77	48	1.8761
0.0353	6.03	50	1.8789
0.0375	6.23	52	1.8803
0.0345	6.48	54	1.8802
0.0346	6.74	56	1.8806

Framework versions

—Transformers 4.37.0.dev0
—Pytorch 2.0.1
—Datasets 2.16.1
—Tokenizers 0.15.0