Full fine-tuning recipe: LoRA on Gemma 2 27B via OpenRLHF, targeting single RTX 3090 (24GB), with data mix and eval plan.
Full fine-tuning recipe: LoRA on Phi-3.5-mini via DeepSpeed, targeting single RTX 3090 (24GB), with data mix and eval plan.
Full fine-tuning recipe: LoRA on DeepSeek-V3 base via Hugging Face TRL, targeting single RTX 3090 (24GB), with data mix and eval plan.
Full fine-tuning recipe: LoRA on Mixtral 8x7B via Megatron-LM, targeting 2x RTX 4090, with data mix and eval plan.
Full fine-tuning recipe: LoRA on Yi 1.5 34B via FSDP, targeting 2x RTX 4090, with data mix and eval plan.
Full fine-tuning recipe: LoRA on Llama 3.3 70B via LLaMA-Factory, targeting AWS g5.12xlarge, with data mix and eval plan.
Full fine-tuning recipe: LoRA on Llama 3.1 70B via Axolotl, targeting AWS g5.12xlarge, with data mix and eval plan.
Full fine-tuning recipe: LoRA on Mistral Small 3 via FSDP, targeting AWS p4d.24xlarge, with data mix and eval plan.
Full fine-tuning recipe: LoRA on Qwen 2.5 7B via LLaMA-Factory, targeting AWS p4d.24xlarge, with data mix and eval plan.
Full fine-tuning recipe: LoRA on Qwen 2.5 32B via Axolotl, targeting AWS p4d.24xlarge, with data mix and eval plan.
Full fine-tuning recipe: LoRA on Qwen 2.5-Coder 7B via LitGPT, targeting Lambda Labs 8xH100, with data mix and eval plan.
Full fine-tuning recipe: LoRA on Gemma 2 27B via torchtune, targeting Lambda Labs 8xH100, with data mix and eval plan.