mistral-7b-base-dpo-run

<!-- This model card has been generated automatically according to the information the Trainer had access to. You should probably proofread and complete it, then remove this comment. -->

<img src="https://raw.githubusercontent.com/OpenAccess-AI-Collective/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>

<details><summary>See axolotl config</summary>

axolotl version: 0.4.0

base_model: mistralai/Mistral-7B-v0.1
base_model_ignore_patterns: []
base_model_config: mistralai/Mistral-7B-v0.1
model_revision:
tokenizer_config:
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer
trust_remote_code: true
tokenizer_use_fast: true
tokenizer_legacy: true
resize_token_embeddings_to_32x: false

is_falcon_derived_model: false
is_llama_derived_model: false
is_mistral_derived_model: true
is_qwen_derived_model: false

model_config:
  rope_scaling:

bnb_config_kwargs:

gptq: false
gptq_groupsize:
gptq_model_v1: false

load_in_8bit: false
load_in_4bit: true

fp16: true

lora_on_cpu: false

rl: dpo
datasets:
  - path: NobodyExistsOnTheInternet/Fixed-gutenberg-dpo-v0.1
    split: train
    type: chatml.intel
  - path: NobodyExistsOnTheInternet/Fixed-Distilabel-intel-orca-dpo-pairs
    split: train
    type: chatml.intel
  - path: NobodyExistsOnTheInternet/ToxicDPOqa
    split: train
    type: chatml.intel
  - path: NobodyExistsOnTheInternet/system-message-DPO
    split: train
    type: chatml.intel
  - path: NobodyExistsOnTheInternet/alpaca-intel-data-dpo
    split: train
    type: chatml.intel
  - path: NobodyExistsOnTheInternet/ToxicDPOqa
    split: train
    type: chatml.intel


chat_template: chatml
default_system_message: Generate a preferable answer.
dataset_prepared_path: data/last_run_prepared
push_dataset_to_hub:
dataset_processes:
dataset_keep_in_memory:
hub_model_id: NobodyExistsOnTheInternet/mistral-7b-base-dpo-run
hub_strategy: every_save
hf_use_auth_token: true
val_set_size: 0
dataset_shard_num:
dataset_shard_idx:

sequence_len: 1024
sample_packing: false
eval_sample_packing:
sample_packing_eff_est:
total_num_tokens:

device_map:
max_memory:

adapter: qlora
lora_model_dir:

lora_r: 32
lora_alpha: 64
lora_dropout: 0.05
lora_target_linear: true
lora_target_module:

lora_modules_to_save:
  - embed_tokens
  - lm_head
lora_fan_in_fan_out:

wandb_project: dpo-hermes-2.5
wandb_entity:
wandb_watch:
wandb_name: 
wandb_run_id:
wandb_log_model:

mlflow_tracking_uri:
mlflow_experiment_name:

output_dir: ./completed-model

torch_compile: true
gradient_accumulation_steps: 4
micro_batch_size: 1
eval_batch_size:
num_epochs: 2
warmup_steps: 100
warmup_ratio:
learning_rate: 0.000001
lr_quadratic_warmup:
logging_steps:
eval_steps:
evals_per_epoch:
save_strategy: steps
save_steps: 1000
saves_per_epoch: 
save_total_limit:
eval_table_size:
eval_max_new_tokens:
eval_causal_lm_metrics:

loss_watchdog_threshold:
loss_watchwatchdog_patience:

train_on_inputs: false
group_by_length: false

gradient_checkpointing: true
gradient_checkpointing_kwargs:
   use_reentrant: false

lr_scheduler:

optimizer: paged_adamw_8bit
weight_decay: 0.01
adam_beta1: 0.95
adam_beta2: 0.999
adam_epsilon: 0.0000001

neftune_noise_alpha: 5

flash_optimum:
xformers_attention:
flash_attention: true
flash_attn_cross_entropy:
flash_attn_rms_norm:
flash_attn_fuse_qkv:
flash_attn_fuse_mlp:
sdp_attention:
s2_attention:
resume_from_checkpoint:
auto_resume_from_checkpoints: false

local_rank:

tokens:

fsdp:
fsdp_config:

deepspeed:

ddp_timeout:
ddp_bucket_cap_mb:
ddp_broadcast_buffers:

torchdistx_path:

pretraining_dataset:

debug:

seed:

</details><br>

mistral-7b-base-dpo-run

This model is a fine-tuned version of mistralai/Mistral-7B-v0.1 on the None dataset.

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-06
• train_batch_size: 1
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 4
• total_train_batch_size: 16
• total_eval_batch_size: 32
• optimizer: Adam with betas=(0.95,0.999) and epsilon=1e-07
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 100
• training_steps: 15031

Training results

Framework versions

• PEFT 0.8.2
• Transformers 4.38.0
• Pytorch 2.2.0+cu121
• Datasets 2.16.1
• Tokenizers 0.15.0