Update model card and embedded training curves

README.md

@@ -7,14 +7,14 @@ tags:
 - global
 - 3b
 - without-metadata
-- continued-pretraining
+- pretraining
 ---
 
 # combined_without_metadata_3b
 
 ## Summary
 
-This repo contains the global combined model at the final 10k-step checkpoint for the metadata localization project. It comes from continued pretraining of a Llama 3.2 base model on the project corpus.
+This repo contains the global combined model at the final 10k-step 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
 
@@ -51,6 +51,22 @@ This repo contains the global combined model at the final 10k-step checkpoint fo
 - `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
+
+![Train Loss](assets/train_loss.png)
+
+### Validation Perplexity
+
+![Validation Perplexity](assets/val_perplexity.png)
+
+### Throughput
+
+![Throughput](assets/tokens_per_sec.png)
+
 ## 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](https://huggingface.co/collections/iamshnoo/metadata-conditioned-llms).
@@ -58,4 +74,4 @@ This model is part of the metadata localization release. Related checkpoints and
 - Project repository: [https://github.com/iamshnoo/metadata_localization](https://github.com/iamshnoo/metadata_localization)
 - Paper: [https://arxiv.org/abs/2601.15236](https://arxiv.org/abs/2601.15236)
 
-Last synced: `2026-04-02 13:49:24 UTC`
+Last synced: `2026-04-02 14:40:46 UTC`