base_model: mistralai/Mistral-7B-v0.1
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer
tokenizer_use_fast: false
resize_token_embeddings_to_32x: false

flash_attention: true
xformers_attention:

load_in_8bit: false
load_in_4bit: false
strict: false

datasets:
  - path: skymizer/Mistral-7B-v0.1-base-tokenized-fineweb-edu-45B-4096
    train_on_split: train
    type: completion

test_datasets:
  - path: skymizer/Mistral-7B-v0.1-base-tokenized-fineweb-edu-test-4K
    split: test
    type: completion

is_preprocess: true
skip_prepare_dataset: true

dataset_prepared_path:

hf_use_auth_token: true
output_dir: /mnt/home/model-team/models/Mistral-7B-v0.1-q-sparse-fineweb-edu-10000steps-4M-bs-table2
resume_from_checkpoint:
auto_resume_from_checkpoints: true

sequence_len: 4096
sample_packing: true
sample_packing_group_size: 100000
sample_packing_bin_size: 200
pad_to_sequence_len: true

eval_sample_packing: false
# eval_causal_lm_metrics: ["perplexity"]

wandb_project: "sparse-tuning-cpt"
wandb_entity:
wandb_watch:
wandb_name: "Mistral-7B-v0.1-q-sparse-fineweb-edu-10000steps-4M-bs-table2"
wandb_log_model:

# global batch size = 2 * 8 * 8 GPUs * 8 Nodes * 4096 = 4M
gradient_accumulation_steps: 8
micro_batch_size: 2
eval_batch_size: 1
max_steps: 10000
optimizer: adamw_torch
learning_rate: 0.00005
lr_scheduler: cosine
cosine_min_lr_ratio: 0.2 
weight_decay: 0.01
adam_beta1: 0.9
adam_beta2: 0.95
adam_eps: 0.000001
max_grad_norm: 2.0

train_on_inputs: false
group_by_length: false
bf16: true
fp16:
tf32: false

hub_model_id: "skymizer/Mistral-7B-v0.1-q-sparse-fineweb-edu-table2"

save_strategy: "steps"
save_steps: 500

gradient_checkpointing: true
gradient_checkpointing_kwargs:
  use_reentrant: false
early_stopping_patience:
local_rank:
logging_steps: 1

warmup_steps: 375
eval_steps: 500
eval_table_size:
debug:
deepspeed: /root/train/axolotl/deepspeed_configs/zero3_bf16.json
fsdp:
fsdp_config:
seed: 42

Mistral-7B-v0.1-q-sparse-fineweb-edu-table2

This model is a fine-tuned version of mistralai/Mistral-7B-v0.1 on an unknown dataset. It achieves the following results on the evaluation set:

Loss: 1.9681

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: 5e-05
train_batch_size: 2
eval_batch_size: 1
seed: 42
distributed_type: multi-GPU
num_devices: 64
gradient_accumulation_steps: 8
total_train_batch_size: 1024
total_eval_batch_size: 64
optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.95) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
lr_scheduler_type: cosine
lr_scheduler_warmup_steps: 375
training_steps: 10000

Training results

Training Loss	Epoch	Step	Validation Loss
11.0878	0.0001	1	11.0613
3.9451	0.0369	500	3.7565
3.4352	0.0738	1000	3.2675
3.149	0.1108	1500	2.9661
2.7926	0.1477	2000	2.6821
2.5117	0.1846	2500	2.4614
2.3734	0.2215	3000	2.3409
2.3038	0.2585	3500	2.2644
2.2295	0.2954	4000	2.2096
2.1764	0.3323	4500	2.1676
2.1645	0.3692	5000	2.1311
2.0971	0.4062	5500	2.1017
2.0761	0.4431	6000	2.0772
2.0447	0.4800	6500	2.0549
2.0683	0.5169	7000	2.0362
2.0181	0.5539	7500	2.0199
2.0193	0.5908	8000	2.0058
1.9909	0.6277	8500	1.9936
2.0054	0.6646	9000	1.9838
1.9898	0.7015	9500	1.9757
1.9738	0.7385	10000	1.9681

Framework versions

Transformers 4.46.3
Pytorch 2.5.1+cu124
Datasets 3.1.0
Tokenizers 0.20.3