VERL Adaptive-G checkpoints

Unified public checkpoint archive for verl_adaptiveG.

Directory convention:

<family>/<model>/<method>/<variant>/<rep>/global_step_<N>/

Each checkpoint directory is a complete Transformers model. Per-item checksums and upload receipts are stored under metadata/items/.

Checkpoint	Repository path	Upload status
Adaptive-G Qwen3-8B Math — Common Sketch G32/B64	`math_rlvr/qwen3-8b/common_sketch/g32_b64_every25/rep4/global_step_200`	complete
Adaptive-G Qwen2.5-Coder-7B — Common Sketch G32/B64	`coding_rlvr/qwen2.5-coder-7b/common_sketch/g32_b64_every25/rep4/global_step_300`	complete
Adaptive-G Qwen3-4B Code — Common Sketch G32/B64	`coding_rlvr/qwen3-4b/common_sketch/g32_b64_every25/rep4/global_step_300`	complete

The archive contains model weights and tokenizer/configuration files. Optimizer and scheduler state are excluded unless a future item is explicitly marked as a resumable training checkpoint.

Downloads last month: -; Downloads are not tracked for this model. How to track

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support