NeuralNovel/Senzu-7B-v0.1-DPO
Embracing a quiet storm ..
Model Details
This model is Senzu-7B-v0.1 a fine-tuned version of mistralai/Mistral-7B-v0.1
DPO Trained on the Neural-DPO dataset.
This model excels at character based roleplay.
Training Parameters
base_model: mistralai/Mistral-7B-v0.1
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer
is_mistral_derived_model: true
load_in_8bit: false
load_in_4bit: false
strict: false
datasets:
- path: practical-dreamer/RPGPT_PublicDomain-alpaca
type: alpaca
format: "[INST] {instruction} [/INST]"
no_input_format: "[INST] {instruction} [/INST]"
datasets:
- path: shuyuej/metamath_gsm8k
type: jeopardy
format: "[INST] {instruction} [/INST]"
no_input_format: "[INST] {instruction} [/INST]"
datasets:
- path: NeuralNovel/Neural-DPO
type:
system_prompt: ""
field_system: system
field_instruction: chosen
field_output: chosen
format: "[INST] {instruction} [/INST]"
no_input_format: "[INST] {instruction} [/INST]"
dataset_prepared_path:
val_set_size: 0.05
output_dir: ./out
sequence_len: 8192
sample_packing: false
pad_to_sequence_len: true
eval_sample_packing: false
wandb_project:
wandb_entity:
wandb_watch:
wandb_name:
wandb_log_model:
gradient_accumulation_steps: 4
micro_batch_size: 2
num_epochs: 1
optimizer: adamw_bnb_8bit
lr_scheduler: cosine
learning_rate: 0.000005
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
warmup_steps: 10
evals_per_epoch: 4
eval_table_size:
eval_max_new_tokens: 128
saves_per_epoch: 0
debug:
deepspeed:
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:
bos_token: "<s>"
eos_token: "</s>"
unk_token: "<unk>"
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 5e-06
- train_batch_size: 2
- eval_batch_size: 2
- seed: 42
- gradient_accumulation_steps: 4
- total_train_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: 1
Training results
| Training Loss | Epoch | Step | Validation Loss |
|---|---|---|---|
| 0.2061 | 0.01 | 1 | 0.3139 |
| 0.0 | 0.25 | 32 | 0.0000 |
| 0.0 | 0.5 | 64 | 0.0010 |
| 0.0 | 0.76 | 96 | 0.0000 |
Framework versions
- Transformers 4.38.0.dev0
- Pytorch 2.2.0+cu121
- Datasets 2.17.1
- Tokenizers 0.15.0
Open LLM Leaderboard Evaluation Results
Detailed results can be found here
| Metric | Value |
|---|---|
| Avg. | 61.90 |
| AI2 Reasoning Challenge (25-Shot) | 66.72 |
| HellaSwag (10-Shot) | 84.34 |
| MMLU (5-Shot) | 62.12 |
| TruthfulQA (0-shot) | 45.29 |
| Winogrande (5-shot) | 79.95 |
| GSM8k (5-shot) | 32.98 |