laion/sera-subset-mixed-10000-axolotl__Qwen3-8B-v8

sera-subset-mixed-10000-axolotl__Qwen3-8B-v8

SFT of Qwen/Qwen3-8B on a 10000-row random mixed subset of ethanlshen/sera-subset (stage1 unresolved + stage2 resolved), trained with axolotl following the upstream SERA recipe.

See baselines/sera/README.md in the open-thoughts/OpenThoughts-Agent repo for the full reproduction details, hyperparameters, and iteration history (this is iteration i9, version v8).

Hyperparameters

• —learning_rate: 1e-5
• —batchsize: 32 (global; micro=1, gradaccum=1, dp=32)
• —num_epochs: 3
• —warmup_steps: 48
• —adambeta1: 0.9, adambeta2: 0.95
• —weight_decay: 0.01
• —sequence_len: 32768
• —chat_template: chatml
• —bf16, deepspeed zero3 (no CPU offload)