--- base_model: - elinas/Llama-3-15B-Instruct-zeroed library_name: transformers tags: - mergekit - merge - finetune datasets: - Chat-Error/Pure-dove-sharegpt license: llama3 --- # Llama-3-15B-Instruct-zeroed-ft-v2 This is a QLoRA **finetune** of a merge of pre-trained language models created using [mergekit](https://github.com/cg123/mergekit). The model is based on a "zeroed" passthrough merge of [Llama-3-15B-Instruct-zeroed](https://huggingface.co/elinas/Llama-3-15B-Instruct-zeroed) This was primarily an experiment to see how a passthrough merge will respond to further finetuning of all LoRA modules. The model was finetuned on **8192 context length** and is likely reliable using RoPE up to 32k. Further finetuning this model or finetuning the [base model](https://huggingface.co/elinas/Llama-3-15B-Instruct-zeroed) on more samples is encouraged. **This will be conducted by myself on the 3rd iteration of this model. Until I receive sufficient feedback on comparison between 8B, finetune v3 will be on hold.** ## Datasets * [Chat-Error/Pure-dove-sharegpt](https://huggingface.co/datasets/Chat-Error/Pure-dove-sharegpt) A small, high quality, curated dataset was used as a PoC / validation on stabilizing the model after the original passthrough merge. ## Finetuning details This is a QLoRA model and all of the LoRA modules were targeted this time to ensure sufficient training before moving on to larger datasets. the first version of this model only targeted **o_proj** and **up_proj** ```yaml lora_target_modules: - gate_proj - down_proj - up_proj - q_proj - v_proj - k_proj - o_proj lora_modules_to_save: - embed_tokens - lm_head ``` The model is coherent even with training the "zeroed" layers plus the additional layers, as this was the recommendation from [Charles Goddard](https://huggingface.co/chargoddard) (mergekit developer) - thank you for sharing the method of merging as well as Toasty Pigeon for bringing it to my attention! ```yaml The following hyperparameters were used during training: - learning_rate: 1e-05 - train_batch_size: 1 - eval_batch_size: 1 - seed: 42 - distributed_type: multi-GPU - num_devices: 3 - total_train_batch_size: 3 - total_eval_batch_size: 3 - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08 - lr_scheduler_type: cosine - lr_scheduler_warmup_steps: 25 - num_epochs: 1 ``` Optimizer `paged_adamw_8bit` and Deepspeed ZeRO 3 was used at a LR of `1e-5` using the cosine scheduler for 1 epoch on 3x3090s taking 4 hours total. **Unsloth** was used for speed and memory savings. Sample packing and padding was disabled to reduce VRAM consumption significantly at the cost of speed. W&B Run Summary ``` wandb: eval/loss 0.90895 wandb: eval/runtime 463.4688 wandb: eval/samples_per_second 0.833 wandb: eval/steps_per_second 0.278 wandb: total_flos 8270790524928.0 wandb: train/epoch 1.0 wandb: train/global_step 1157 wandb: train/grad_norm 7.3847 wandb: train/learning_rate 0.0 wandb: train/loss 0.8702 wandb: train_loss 0.87814 wandb: train_runtime 16425.2713 wandb: train_samples_per_second 0.211 wandb: train_steps_per_second 0.07 ``` ### Framework versions - PEFT 0.10.0 - Transformers 4.40.2 - Pytorch 2.3.0+cu121 - Datasets 2.19.1 - Tokenizers 0.19.1 ## Model Evaluation TBD If you have any questions or comments on the model, feel free to open a discussion in the community tab. [Built with Axolotl](https://github.com/OpenAccess-AI-Collective/axolotl)