tkwiecinski/amr-fma-Apertus-8B-Instruct-2509-lora_sft-em_insecure-e2_em_ablation-s43

amr-fma training run.

• Method: lora_sft
• Base model: swiss-ai/Apertus-8B-Instruct-2509
• Dataset: insecure (slug: em_insecure)
• Seed: 43
• Git commit: 6b9a7d1eb9d81484964a71ad6829945e7680efad
• Exp name: e2_em_ablation
• WandB run: tl8rmgvi

Tags

• phase:P1
• domain:code

Checkpoints (branches)

• step 1 → revision step-00001
• step 3 → revision step-00003
• step 5 → revision step-00005
• step 10 → revision step-00010
• step 17 → revision step-00017
• step 31 → revision step-00031
• step 57 → revision step-00057

Pin a specific checkpoint with revision=... in AutoModelForCausalLM.from_pretrained / PeftModel.from_pretrained.

Hyperparameter sections

checkpointing, dataset, evaluation, final_adapter_path, lora, model, optimization, prompt_style, runtime, sdpo, sequence, total_steps