File size: 2,277 Bytes
5df12ed a9146db 5df12ed a9146db a0b1188 f8f020a a9146db e215834 5df12ed a9146db 5df12ed a9146db 5df12ed a9146db 5df12ed a9146db 5df12ed a9146db 5df12ed a9146db 5df12ed a9146db 5df12ed |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 |
---
library_name: transformers
tags:
- mergekit
datasets:
- arcee-ai/sec-data-full
---
# arcee-ai/Mistral-7B-Instruct-v0.2-expanded-sec-1.6B-tokens
This model was trained for a single epoch with 1.6B SEC data in Llama-pro style
### Model Description
We only trained the newly added blocks as in the Llama pro paper while keeping every other layer frozen.
- **Finetuned from model:** arcee-ai/Mistral-7B-Instruct-v0.2-expanded
### Model Sources
<!-- Provide the basic links for the model. -->
- **Repository:** Used huggingface/alignment-handbook - https://github.com/huggingface/alignment-handbook/tree/main/recipes/gpt2-nl/cpt
- **Paper :** LLAMA PRO: Progressive LLaMA with Block Expansion [https://arxiv.org/pdf/2401.02415.pdf]
## Uses
Can use with SEC data.
#### Training Hyperparameters
Trained with p5.48xlarge GPU (8 x 80GB).
```
# Model arguments
model_name_or_path: arcee-ai/Mistral-7B-Instruct-v0.2-expanded
model_revision: main
torch_dtype: bfloat16
# Data training arguments
dataset_mixer:
arcee-ai/sec-data-full: 1.0
dataset_splits:
- train
preprocessing_num_workers: 12
# SFT trainer config
bf16: true
do_eval: False
evaluation_strategy: "no"
gradient_accumulation_steps: 32
gradient_checkpointing: true
gradient_checkpointing_kwargs:
use_reentrant: False
hub_model_id: arcee-ai/mistral-instruct-v2-sec-expanded
hub_strategy: every_save
learning_rate: 2.0e-05
log_level: info
logging_steps: 1
logging_strategy: steps
lr_scheduler_type: cosine
max_seq_length: 2048
max_steps: -1
num_train_epochs: 1
output_dir: data/mistral-instruct-v2-sec-expanded-new
overwrite_output_dir: true
per_device_eval_batch_size: 1
per_device_train_batch_size: 16
push_to_hub: true
remove_unused_columns: true
report_to:
- wandb
save_strategy: "steps"
save_steps: 100
save_total_limit: 1
seed: 42
warmup_ratio: 0.01
# ACCELERATE_LOG_LEVEL=info accelerate launch --config_file recipes/accelerate_configs/multi_gpu.yaml --num_processes=8 scripts/run_cpt.py recipes/gpt2-nl/cpt/config_full.yaml
```
#### Loss Curves
|