DatedGPT-2023 (base)
DatedGPT is a family of point-in-time language models: each vintage is trained only on data available up to its cutoff date, making it suitable for lookahead-bias-free prediction and point-in-time analysis.
This is the base (pretrained) model with data up to 2023 — no
instruction tuning. For the instruction-tuned variants, see the
datedgpt-instruct-* repositories in this organization.
| Property | Value |
|---|---|
| Architecture | LlamaForCausalLM |
| Parameters | ~1.3 B |
| Context length | 2048 |
| Vocab | 32,000 (SentencePiece) |
| Precision | bfloat16 |
| Data vintage | 2023 |
Usage
import torch
from transformers import AutoTokenizer, AutoModelForCausalLM
repo_id = "datedgpt/datedgpt-2023-base"
tokenizer = AutoTokenizer.from_pretrained(repo_id)
model = AutoModelForCausalLM.from_pretrained(repo_id, torch_dtype=torch.bfloat16, device_map="auto")
inputs = tokenizer("The stock market in 2023", return_tensors="pt").to(model.device)
output = model.generate(**inputs, max_new_tokens=64, use_cache=True)
print(tokenizer.decode(output[0], skip_special_tokens=True))
Limitations
- Base model: completions only, no chat/instruction following.
- Knowledge limited to the 2023 data vintage.
- No RLHF or safety tuning.