Full fine-tuning recipe: QLoRA (4-bit) on Yi 1.5 34B via torchtune, targeting single RTX 4090 (24GB), with data mix and eval plan.
Full fine-tuning recipe: QLoRA (4-bit) on Yi 1.5 34B via Megatron-LM, targeting AWS p4d.24xlarge, with data mix and eval plan.
Full fine-tuning recipe: QLoRA (4-bit) on Yi 1.5 34B via torchtune, targeting 4x A100 40GB, with data mix and eval plan.
Full fine-tuning recipe: QLoRA (4-bit) on DeepSeek-V3 base via FSDP, targeting single RTX 3090 (24GB), with data mix and eval plan.