base_model: mistralai/Mistral-7B-v0.1
model_type: MistralForCausalLM
tokenizer_type: LlamaTokenizer

load_in_8bit: true
load_in_4bit: false
strict: false

datasets:
  # - path: mhenrichsen/alpaca_2k_test
  # - path: /home/yujia/home/CN_Hateful/train_toxiCN.json
  # - path: /home/yujia/home/CN_Hateful/train_toxiCN_cn.json
  - path: /home/yujia/home/CN_Hateful/train.json
    ds_type: json
    type: alpaca
dataset_prepared_path: last_run_prepared
val_set_size: 0.1
# output_dir: /home/yujia/home/CN_Hateful/trained_models/mistral/toxi/1e-5/
# output_dir: /home/yujia/home/CN_Hateful/trained_models/mistral/CN/toxi/1e-5/
output_dir: /home/yujia/home/CN_Hateful/trained_models/mistral/cold/3e-4/



adapter: lora
lora_model_dir:

sequence_len: 256
sample_packing: true
pad_to_sequence_len: true

lora_r: 32
lora_alpha: 16
lora_dropout: 0.05
lora_target_linear: true
lora_fan_in_fan_out:
lora_target_modules:
  - gate_proj
  - down_proj
  - up_proj
  - q_proj
  - v_proj
  - k_proj
  - o_proj

wandb_project:
wandb_entity:
wandb_watch:
wandb_name:
wandb_log_model:

gradient_accumulation_steps: 8
micro_batch_size: 4
num_epochs: 3
optimizer: adamw_bnb_8bit
lr_scheduler: cosine
learning_rate: 0.0003

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

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

loss_watchdog_threshold: 5.0
loss_watchdog_patience: 3

warmup_steps: 10
evals_per_epoch: 4
eval_table_size:
eval_max_new_tokens: 128
saves_per_epoch: 1
debug:
deepspeed:
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:

home/yujia/home/CN_Hateful/trained_models/mistral/cold/3e-4/

This model is a fine-tuned version of mistralai/Mistral-7B-v0.1 on the None dataset. It achieves the following results on the evaluation set:

Loss: 0.0406

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: 0.0003
train_batch_size: 4
eval_batch_size: 4
seed: 42
distributed_type: multi-GPU
num_devices: 2
gradient_accumulation_steps: 8
total_train_batch_size: 64
total_eval_batch_size: 8
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
2.9655	0.01	1	2.9710
0.0451	0.25	46	0.0619
0.0541	0.5	92	0.0392
0.0353	0.75	138	0.0345
0.0249	1.0	184	0.0315
0.0259	1.23	230	0.0329
0.0238	1.48	276	0.0309
0.019	1.73	322	0.0305
0.0173	1.97	368	0.0313
0.0051	2.21	414	0.0369
0.0093	2.46	460	0.0436
0.0023	2.71	506	0.0407
0.003	2.95	552	0.0406

Framework versions

PEFT 0.10.0
Transformers 4.40.0.dev0
Pytorch 2.2.1+cu121
Datasets 2.18.0
Tokenizers 0.15.0

yujia23/axolotl-mistral-cold-3e-4-lora

home/yujia/home/CN_Hateful/trained_models/mistral/cold/3e-4/

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

yujia23/
axolotl-mistral-cold-3e-4-lora