You can deploy this model and start earning money today!

MattBou00/
mq028hjz-rlhf-checkpoint-pythia-1b-irl-epoch-80

Name: mq028hjz-rlhf-checkpoint-pythia-1b-irl-epoch-80
Availability: InStock

safetensors

gpt_neox

Views

No views yet

mq028hjz-rlhf-checkpoint-pythia-1b-irl-epoch-80

This is a RLHF model checkpoint trained at epoch 80.

Model Information

Base Model: EleutherAI/pythia-1b
Reward Type: irl
Dataset: allenai/real-toxicity-prompts
Training Epoch: 80

IRL Configuration

Likelihood Type: bradley_terry
Normalization Strategy: none
IRL Artifact: matthieubou-imperial-college-london/bayes_irl_vi/posterior_bradley_terry_05megofd:v0
Use Raw Score: True

Usage

This checkpoint can be loaded using the HuggingFace Transformers library:

python

from transformers import AutoModelForCausalLM
from trl import AutoModelForCausalLMWithValueHead

# Load the checkpoint
model = AutoModelForCausalLMWithValueHead.from_pretrained("MattBou00/mq028hjz-rlhf-checkpoint-pythia-1b-irl-epoch-80")

Training Configuration

The training configuration is saved in training_config.yaml.

language: en tags:

rlhf
checkpoint
irl
pythia-1b library_name: transformers pipeline_tag: text-generation