adrian27513/mistral_math_genie_2

See axolotl config

axolotl version: 0.8.1

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

load_in_8bit: false
load_in_4bit: false
strict: false

torch_compile:  true
torch_compile_backend: inductor

lora_mlp_kernel: true
lora_qkv_kernel: true
lora_o_kernel: true

datasets:
  - path: ./math_genie/patch/train
    type:
      system_prompt: ""
      field_system: system
      field_instruction: question
      field_output: answer
      format: "Question:{instruction}\nAnswer:"
      no_input_format: "Question:{instruction}\nAnswer:"
    ds_type: arrow 

val_set_size: 0.05

output_dir: ./peft_output/mistral7b-math_genie_patch

sequence_len: 2048
sample_packing: true
pad_to_sequence_len: true
eval_sample_packing: false

adapter: lora
lora_model_dir:

lora_r: 8
lora_alpha: 16
lora_dropout: 0.15
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: 1
micro_batch_size: 16
num_epochs: 4
optimizer: adamw_torch_fused
learning_rate: 2e-4
lr_scheduler: cosine
lr_scheduler_kwargs:
cosine_min_lr_ratio: 0.1
cosine_constant_lr_ratio: 0.8

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

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

loss_watchdog_threshold: 20.0
loss_watchdog_patience: 5

warmup_ratio: 0.1
evals_per_epoch: 1
saves_per_epoch: 2
save_strategy: best
debug:
deepspeed:
weight_decay: 0.02
fsdp:
fsdp_config:
special_tokens:

plugins:
  - axolotl.integrations.liger.LigerPlugin
liger_rope: true
liger_rms_norm: true
liger_glu_activation: true
liger_layer_norm: true
liger_fused_linear_cross_entropy: true

peft_output/mistral7b-math_genie_patch

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: 0.2911

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: 0.0002
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• total_train_batch_size: 128
• total_eval_batch_size: 128
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 7
• num_epochs: 4.0

Training results

Training Loss	Epoch	Step	Validation Loss
0.6507	0.0526	1	0.6676
0.3424	1.0	19	0.3352
0.2915	2.0	38	0.3036
0.2815	3.0	57	0.2944
0.2869	4.0	76	0.2911

Framework versions

• PEFT 0.15.1
• Transformers 4.51.0
• Pytorch 2.6.0+cu126
• Datasets 3.5.0
• Tokenizers 0.21.1