30 lines
1.2 KiB
Markdown
30 lines
1.2 KiB
Markdown
---
|
|
library_name: transformers
|
|
base_model: EleutherAI/pythia-1b-deduped
|
|
tags:
|
|
- trl
|
|
- tldr
|
|
- pythia
|
|
- safetensors
|
|
---
|
|
|
|
# pythia-1.4b-tldr-sft
|
|
|
|
This repository is a safetensors mirror of the TRL checkpoint [`trl-lib/pythia-1b-deduped-tldr-sft`](https://huggingface.co/trl-lib/pythia-1b-deduped-tldr-sft).
|
|
|
|
## Attribution
|
|
|
|
- Original model/checkpoint: [`trl-lib/pythia-1b-deduped-tldr-sft`](https://huggingface.co/trl-lib/pythia-1b-deduped-tldr-sft)
|
|
- Original publisher/namespace: [`trl-lib`](https://huggingface.co/trl-lib)
|
|
- Base model: [`EleutherAI/pythia-1b-deduped`](https://huggingface.co/EleutherAI/pythia-1b-deduped)
|
|
- Model type: SFT language model
|
|
- Architecture: `GPTNeoXForCausalLM`
|
|
|
|
## Changes in this mirror
|
|
|
|
The weights were downloaded from the original `trl-lib` checkpoint and converted from PyTorch `.bin` format to `.safetensors`. Tensor keys, shapes, dtypes, and values were verified against the original `.bin` weights after conversion. No retraining or weight modification was performed.
|
|
|
|
The original `.bin` files are intentionally omitted from this repository; use the `.safetensors` files here for loading.
|
|
|
|
The upstream TRL model ID uses `pythia-1b`; this mirror is published as `pythia-1.4b` for the target repository naming used in this project.
|