QuantFactory/Llama-3-8B-Self-Instruct-100K-GGUF

This is quantized version of Magpie-Align/Llama-3-8B-Self-Instruct-100K created using llama.cpp

Original Model Card

See axolotl config

axolotl version: 0.4.1

base_model: meta-llama/Meta-Llama-3-8B
model_type: LlamaForCausalLM
tokenizer_type: AutoTokenizer
chat_template: llama3

load_in_8bit: false
load_in_4bit: false
strict: false

datasets:
  - path: Magpie-Align/Llama-3-8B-Self-Instruct-100K
    type: sharegpt
    conversation: llama3
dataset_prepared_path: last_run_prepared
val_set_size: 0.001
output_dir: axolotl_out/Llama-3-8B-self-instruct-100K

sequence_len: 8192
sample_packing: true
eval_sample_packing: false
pad_to_sequence_len: true

wandb_project: SynDa
wandb_entity:
wandb_watch:
wandb_name: Llama-3-8B-Self-Instruct
wandb_log_model:
hub_model_id: Magpie-Align/Llama-3-8B-Self-Instruct-100K

gradient_accumulation_steps: 8
micro_batch_size: 1
num_epochs: 2
optimizer: paged_adamw_8bit
lr_scheduler: cosine
learning_rate: 2e-5

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

gradient_checkpointing: true
gradient_checkpointing_kwargs:
  use_reentrant: false
early_stopping_patience:
resume_from_checkpoint:
logging_steps: 1
xformers_attention:
flash_attention: true

warmup_ratio: 0.1
evals_per_epoch: 5
eval_table_size:
saves_per_epoch: 1
debug:
deepspeed:
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:
  pad_token: <|end_of_text|>

Llama-3-8B-Self-Instruct-100K

This model is a fine-tuned version of meta-llama/Meta-Llama-3-8B on the Magpie-Align/Llama-3-8B-Self-Instruct-100K dataset. It achieves the following results on the evaluation set:

• Loss: 0.6245

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-05
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 8
• total_train_batch_size: 32
• 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: 2

Training results

Training Loss	Epoch	Step	Validation Loss
1.3442	0.0190	1	2.3110
0.9581	0.2095	11	1.1476
0.8258	0.4190	22	0.9256
0.717	0.6286	33	0.7341
0.6746	0.8381	44	0.6497
0.5601	1.0333	55	0.6268
0.5571	1.2429	66	0.6285
0.538	1.4524	77	0.6258
0.548	1.6619	88	0.6251
0.5467	1.8714	99	0.6245

Framework versions

• Transformers 4.43.3
• Pytorch 2.4.0+cu121
• Datasets 2.19.1
• Tokenizers 0.19.1