axolotl version: 0.4.0
adapter: null
base_model: NousResearch/Nous-Hermes-2-Mistral-7B-DPO
batch_size: 2
bf16: auto
dataset_prepared_path: null
datasets:
- ds_type: json
path: /workspace/data.jsonl
type: context_qa.load_v2
debug: null
deepspeed: null
early_stopping_patience: null
evals_per_epoch: 4
flash_attention: null
fp16: null
fsdp: null
fsdp_config: null
gptq_groupsize: null
gptq_model_v1: null
gradient_checkpointing: true
group_by_length: false
learning_rate: 1.0e-05
local_rank: null
logging_steps: 1
lora_alpha: 32
lora_dropout: 0.2
lora_fan_in_fan_out: null
lora_model_dir: null
lora_r: 64
lora_target_linear: true
lora_target_modules: null
lr_scheduler: cosine
max_packed_sequence_len: null
micro_batch_size: 1
model_config:
output_router_logits: true
model_type: MistralForCausalLM
num_epochs: 4
optimizer: adamw_bnb_8bit
output_dir: /workspace/out-mistral-2B
resume_from_checkpoint: null
saves_per_epoch: 1
sequence_len: 2048
special_tokens:
bos_token: <s>
eos_token: <|im_end|>
pad_token: </s>
tf32: true
tokenizer_type: LlamaTokenizer
torchdistx_path: null
train_on_inputs: false
trust_remote_code: true
val_set_size: 0.05
wandb_entity: null
wandb_log_model: Nous-Hermes-2-Mistral-7B-DPO
wandb_name: mistral
wandb_project: Ultron-llama
wandb_watch: null
warmup_steps: 40
weight_decay: 0.0
xformers_attention: true
workspace/out-mistral-2B
This model is a fine-tuned version of NousResearch/Nous-Hermes-2-Mistral-7B-DPO on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.5036
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: 1e-05
- train_batch_size: 1
- eval_batch_size: 1
- seed: 42
- distributed_type: multi-GPU
- num_devices: 2
- gradient_accumulation_steps: 2
- total_train_batch_size: 4
- total_eval_batch_size: 2
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: cosine
- lr_scheduler_warmup_steps: 40
- num_epochs: 4
Training results
| Training Loss | Epoch | Step | Validation Loss |
|---|---|---|---|
| 0.6411 | 0.02 | 1 | 0.4803 |
| 0.5321 | 0.26 | 11 | 0.3867 |
| 0.4077 | 0.51 | 22 | 0.3591 |
| 0.4455 | 0.77 | 33 | 0.3995 |
| 0.2921 | 1.02 | 44 | 0.4368 |
| 0.3459 | 1.28 | 55 | 0.4884 |
| 0.2768 | 1.53 | 66 | 0.4978 |
| 0.4168 | 1.79 | 77 | 0.4808 |
| 0.14 | 2.05 | 88 | 0.4547 |
| 0.1132 | 2.3 | 99 | 0.4856 |
| 0.1055 | 2.56 | 110 | 0.4916 |
| 0.1385 | 2.81 | 121 | 0.4783 |
| 0.0455 | 3.07 | 132 | 0.4677 |
| 0.0211 | 3.33 | 143 | 0.4892 |
| 0.0236 | 3.58 | 154 | 0.5016 |
| 0.009 | 3.84 | 165 | 0.5036 |
Framework versions
- Transformers 4.39.0.dev0
- Pytorch 2.1.2+cu121
- Datasets 2.18.0
- Tokenizers 0.15.0