This repo contains the global combined model exported from the 8k checkpoint for the metadata localization project. It was trained from scratch on the project corpus, using the Llama 3.2 tokenizer and vocabulary.

Variant Metadata

Stage: pretrain
Family: global
Size: 3b
Metadata condition: without_metadata
Checkpoint export: 8k
Base model lineage: Trained from scratch; tokenizer/vocabulary from meta-llama/Llama-3.2-1B

Weights & Biases Provenance

Run name: 31/03/2026_15:42:36_combined_without_metadata_3b
Internal run URL: https://wandb.ai/iamshnoo/nanotron/runs/gwgpbgnj
Note: the Weights & Biases workspace is private; public readers should use the summarized metrics and configuration below.
State: finished
Runtime: 5h 13m 37s

Run Summary

KPI/train_lm_loss: 2.1208
KPI/train_perplexity: 8.338
KPI/val_loss: 2.2234
KPI/val_perplexity: 9.2383
KPI/consumed_tokens/train: 41,943,040,000
_step: 10,000

Training Configuration

train_steps: 10,000
sequence_length: 2,048
micro_batch_size: 8
batch_accumulation_per_replica: 64
learning_rate: 0.0003
min_decay_lr: 0
checkpoint_interval: 100

Training Curves

Static plots below were exported from the private Weights & Biases run and embedded here for public access.

Train Loss

Validation Perplexity

Throughput

Project Context

This model is part of the metadata localization release. Related checkpoints and variants are grouped in the public Hugging Face collection Metadata Conditioned LLMs.

Training data source: News on the Web (NOW) Corpus
Project repository: https://github.com/iamshnoo/metadata_localization
Paper: https://arxiv.org/abs/2601.15236

Last synced: 2026-04-02 14:40:40 UTC

iamshnoo/combined_without_metadata_3b_step8k

combined_without_metadata_3b_step8k

Summary