---
base_model: Medixspace/Mistral-Nemo-Instruct-2407-freeze-with-TNM
library_name: peft
---

Checkpoint: /media/mlazure4775847383/storage/LM_saves_Mistral/Mistral-Nemo-Instruct-2407/lora/train_2024-11-05-09-59-46/checkpoint-2103

NB! The tokenizer in this repo has been changed for the base model's tokenizer. This is needed for predibase run. The original tokenizer is in storage.

Training args:
```
bf16: true
cutoff_len: 25000
dataset: train_data_checklist,val_data_checklist
dataset_dir: data
ddp_timeout: 180000000
deepspeed: cache/ds_z3_offload_config.json
do_train: true
finetuning_type: lora
flash_attn: fa2
gradient_accumulation_steps: 8
include_num_input_tokens_seen: true
learning_rate: 5.0e-05
logging_steps: 250
lora_alpha: 16
lora_dropout: 0
lora_rank: 8
lora_target: all
lr_scheduler_type: cosine
max_grad_norm: 1.0
max_samples: 100000
model_name_or_path: Medixspace/Mistral-Nemo-Instruct-2407-freeze-with-TNM
num_train_epochs: 3.0
optim: adamw_torch
output_dir: saves/Mistral-Nemo-Instruct-2407/lora/train_2024-11-05-09-59-46
packing: false
per_device_train_batch_size: 1
plot_loss: true
preprocessing_num_workers: 16
report_to: none
save_steps: 500
stage: sft
template: mistral
warmup_steps: 0
```

Best inference args:
```
temperature=0.9
top-p=0.5
```