ShenaoZhang/0.001_idpo_iter_2

TEXT GENERATIONConcurrency Cost:1Model Size:7BQuant:FP8Ctx Length:8kPublished:Apr 5, 2024License:mitArchitecture:Transformer Open Weights Cold

The ShenaoZhang/0.001_idpo_iter_2 model is a fine-tuned iteration building upon ShenaoZhang/0.001_idpo_iter_1, developed by ShenaoZhang. It was trained using specific hyperparameters including a learning rate of 5e-07 and a total batch size of 128 over 1 epoch. This model is part of an iterative development process, with its primary differentiation stemming from its fine-tuning on the ShenaoZhang/0.001_idpo_dataset. Its specific capabilities and intended uses require further information for precise application.

Loading preview...

Model Overview

ShenaoZhang/0.001_idpo_iter_2 is an iteratively fine-tuned language model developed by ShenaoZhang. It is a direct successor to ShenaoZhang/0.001_idpo_iter_1, having been fine-tuned on the ShenaoZhang/0.001_idpo_dataset.

Training Details

The model underwent a single training epoch with a learning rate of 5e-07. Key training hyperparameters included:

  • Optimizer: Adam with betas=(0.9, 0.999) and epsilon=1e-08
  • Batch Sizes: train_batch_size of 8, eval_batch_size of 8, leading to a total_train_batch_size of 128 and total_eval_batch_size of 64 (with 8 devices and 2 gradient accumulation steps).
  • Scheduler: Cosine learning rate scheduler with a warmup ratio of 0.1.
  • Seed: 42 for reproducibility.

Framework Versions

The training environment utilized:

  • Transformers 4.36.2
  • Pytorch 2.1.2+cu121
  • Datasets 2.14.6
  • Tokenizers 0.15.2

Current Status

Further information regarding the model's specific capabilities, intended uses, limitations, and evaluation results is currently pending. Users are advised that detailed performance metrics and specific application guidance are not yet available.