Downloads · 30 days
0
dureduck/gemma4-26b-a4b-gsm8k-synthetic-cot6144-sft-20260916
gemma4-26b-a4b-gsm8k-synthetic-cot6144-sft-20260916 is a machine learning model from dureduck. Use it for the machine learning task on the model card, and read the license before you ship it in a product. It is set up for maxtext.
Full native Orbax/Tunix training checkpoints for Jobman experiment gemma4-26b-a4b-gsm8k-synthetic-cot6144-sft-20260916, at optimizer updates 80, 160, and 233. Each published checkpoint has a checkpoint-STEP tag and re…
Downloads · 30 days
0
Access
Public
Updated Sep 19, 2026
Repo size
370 GB
Likes
0
Public
Click a slice to open those files.
Other370 GB · 100%
From the Hugging Face model README
Full native Orbax/Tunix training checkpoints for Jobman experiment gemma4-26b-a4b-gsm8k-synthetic-cot6144-sft-20260916, at optimizer updates 80, 160, and 233. Each published checkpoint has a checkpoint-STEP tag and resides in checkpoints/STEP/.
These are native training checkpoints, including model parameters and optimizer state, not Transformers safetensors. Preserve the entire step directory when restoring. resume-state.json records data-progress information; publication-manifest.json records source generations and payload hashes. Public training data is referenced by revision and hash rather than bundled. Credentials are excluded.
Base model revision: 24548b62aa021d562695c04aaf7758a1ea47990b. MaxText revision: cc79eb8b0172849875f9dd767946113dc9c40d28. Sequence length 6144; one training epoch; effective batch32; BF16 weights/optimizer; shuffle seed42.
The final native save omitted custom metadata; the final checkpoint sidecar explicitly records this and binds its optimizer/data progress to the immutable committed step233 snapshot. Upload integrity verification is distinct from verification by successful training restore.