mlfoundations-dev_-_gemma-oh-preferences-4bits

Quantization made by Richard Erkhov.

Github

Discord

Request more models

gemma-oh-preferences - bnb 4bits

• Model creator: https://huggingface.co/mlfoundations-dev/
• Original model: https://huggingface.co/mlfoundations-dev/gemma-oh-preferences/

Original model description:

library_name: transformers license: gemma base_model: google/gemma-2-9b-it tags:

• llama-factory
• full
• trl
• dpo
• llama-factory
• generated_from_trainer model-index:
• name: gemma-oh-preferences results: []

---

<!-- 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. -->

gemma-oh-preferences

This model is a fine-tuned version of google/gemma-2-9b-it on the cdc0b2d9-493b-4cb1-87e8-8fb1e3f4b247 dataset. It achieves the following results on the evaluation set:

• Loss: 3.9424
• Rewards/chosen: -46.0706
• Rewards/rejected: -47.7080
• Rewards/accuracies: 0.6460
• Rewards/margins: 1.6374
• Logps/rejected: -4.7708
• Logps/chosen: -4.6071
• Logits/rejected: 14.6998
• Logits/chosen: 14.4715

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: 8e-07
• train_batch_size: 2
• eval_batch_size: 2
• seed: 42
• distributed_type: multi-GPU
• num_devices: 32
• gradient_accumulation_steps: 2
• total_train_batch_size: 128
• total_eval_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 1.0

Training results

Framework versions

• Transformers 4.45.2
• Pytorch 2.3.0
• Datasets 2.21.0
• Tokenizers 0.20.2