|
Download README.md from NextTransformer/Circle-Loop-49B-Tokens: direct link, hf CLI and curl.
- Browser
- Download file 2.37 kB
-
https://huggingface.co/NextTransformer/Circle-Loop-49B-Tokens/resolve/main/README.md
- Command line
-
hf download hf://NextTransformer/Circle-Loop-49B-Tokens/README.md
-
curl -L -o README.md https://huggingface.co/NextTransformer/Circle-Loop-49B-Tokens/resolve/main/README.md
2.37 kB
metadata
library_name: pytorch
tags:
- nexttransformer
- looped
- causal-lm
- pretrained
- safetensors
pipeline_tag: text-generation
NextTransformer Circle Loop
After the first circuit, every block becomes an exit.
This is the final Circle Loop checkpoint in the NextTransformer 49.152B-token architecture comparison. It has 927,992,523 parameters and a 2,048-token context. The model was pretrained on ClimbMix for 46,875 updates. The 49B in the repository name refers to training tokens.
| Final evaluation | Result |
|---|---|
| Validation BPB | 0.6912 |
| CORE-22 (×100) | 28.97 |
Load the weights
Install the tested NextTransformer implementation and the safetensors reader in a Python 3.10–3.13 environment.
pip install "nexttransformer @ git+https://github.com/Interactive-Active-AI/NextTransformer.git@7aa267a88b45e4a19a9739ccb4ba3f196d7aacc0" safetensors
The example encodes a prompt with the checkpoint's tokenizer and prints its most likely next token.
import base64
import json
from pathlib import Path
import tiktoken
import torch
from huggingface_hub import snapshot_download
from safetensors.torch import load_file
from nexttransformer.models import build_model, config_from_dict
root = Path(snapshot_download("NextTransformer/Circle-Loop-49B-Tokens"))
config = config_from_dict(json.loads((root / "model_config.json").read_text()))
model = build_model(config, device="cpu")
model.load_state_dict(load_file(root / "model.safetensors"), strict=True)
if hasattr(model, "set_step"):
model.set_step(46875)
model.eval()
spec = json.loads((root / "tokenizer.tiktoken.json").read_text())
encoding = tiktoken.Encoding(
name=spec["name"],
pat_str=spec["pat_str"],
mergeable_ranks={base64.b64decode(b): rank for b, rank in spec["mergeable_ranks"]},
special_tokens=spec["special_tokens"],
)
prompt = [encoding.encode_single_token("<|bos|>")] + encoding.encode_ordinary("The sky is")
with torch.inference_mode():
logits = model(torch.tensor([prompt]))
print(encoding.decode([int(logits[0, -1].argmax())]))