Full fine-tuning recipe: IPO on Mistral Nemo 12B via Megatron-LM, targeting 2x A100 80GB, with data mix and eval plan.
Full fine-tuning recipe: IPO on Qwen 2.5 32B via FSDP, targeting 2x A100 80GB, with data mix and eval plan.
Full fine-tuning recipe: IPO on Qwen 2.5-Coder 7B via LLaMA-Factory, targeting 4x A100 40GB, with data mix and eval plan.
Full fine-tuning recipe: IPO on Gemma 2 27B via Axolotl, targeting 4x A100 40GB, with data mix and eval plan.
Full fine-tuning recipe: IPO on Phi-3.5-mini via LitGPT, targeting 8x H100, with data mix and eval plan.
Full fine-tuning recipe: IPO on DeepSeek-V3 base via torchtune, targeting 8x H100, with data mix and eval plan.
Full fine-tuning recipe: IPO on Mixtral 8x7B via Axolotl, targeting 8x H100, with data mix and eval plan.
Full fine-tuning recipe: IPO on Yi 1.5 34B via LitGPT, targeting single RTX 4090 (24GB), with data mix and eval plan.
Full fine-tuning recipe: IPO on Llama 3.3 70B via torchtune, targeting single RTX 4090 (24GB), with data mix and eval plan.
Full fine-tuning recipe: IPO on Llama 3.1 8B via Unsloth, targeting single RTX 3090 (24GB), with data mix and eval plan.
Full fine-tuning recipe: IPO on Mixtral 8x22B via LLaMA-Factory, targeting AWS p4d.24xlarge, with data mix and eval plan.
Full fine-tuning recipe: IPO on Mixtral 8x22B via OpenRLHF, targeting 4x A100 40GB, with data mix and eval plan.