GGML converted versions of EleutherAI's GPT-J model
Description
GPT-J 6B is a transformer model trained using Ben Wang's Mesh Transformer JAX. "GPT-J" refers to the class of model, while "6B" represents the number of trainable parameters.
<figure>
| Hyperparameter | Value |
|---|---|
| \(n_{parameters}\) | 6053381344 |
| \(n_{layers}\) | 28* |
| \(d_{model}\) | 4096 |
| \(d_{ff}\) | 16384 |
| \(n_{heads}\) | 16 |
| \(d_{head}\) | 256 |
| \(n_{ctx}\) | 2048 |
| \(n_{vocab}\) | 50257/50400† (same tokenizer as GPT-2/3) |
| Positional Encoding | Rotary Position Embedding (RoPE) |
| RoPE Dimensions | 64 |
| <figcaption><p><strong>*</strong> Each layer consists of one feedforward block and one self attention block.</p> | |
| <p><strong>†</strong> Although the embedding matrix has a size of 50400, only 50257 entries are used by the GPT-2 tokenizer.</p></figcaption></figure> |
The model consists of 28 layers with a model dimension of 4096, and a feedforward dimension of 16384. The model dimension is split into 16 heads, each with a dimension of 256. Rotary Position Embedding (RoPE) is applied to 64 dimensions of each head. The model is trained with a tokenization vocabulary of 50257, using the same set of BPEs as GPT-2/GPT-3.
Converted Models
| Name | Based on | Type | Container | GGML Version |
|---|---|---|---|---|
| gpt-j-6b-f16.bin | EleutherAI/gpt-j-6b | F16 | GGML | V3 |
| gpt-j-6b-q4_0.bin | EleutherAI/gpt-j-6b | Q4_0 | GGML | V3 |
| gpt-j-6b-q4_0-ggjt.bin | EleutherAI/gpt-j-6b | Q4_0 | GGJT | V3 |
| gpt-j-6b-q5_1.bin | EleutherAI/gpt-j-6b | Q5_1 | GGML | V3 |
| gpt-j-6b-q5_1-ggjt.bin | EleutherAI/gpt-j-6b | Q5_1 | GGJT | V3 |
Usage
Python via llm-rs:
Installation
Via pip: pip install llm-rs
Run inference
from llm_rs import AutoModel
#Load the model, define any model you like from the list above as the `model_file`
model = AutoModel.from_pretrained("rustformers/gpt-j-ggml",model_file="gpt-j-6b-q4_0-ggjt.bin")
#Generate
print(model.generate("The meaning of life is"))
Rust via Rustformers/llm:
Installation
git clone --recurse-submodules https://github.com/rustformers/llm.git
cd llm
cargo build --release
Run inference
cargo run --release -- gptj infer -m path/to/model.bin -p "Tell me how cool the Rust programming language is:"
