update README and add config file

README.md

@@ -9,6 +9,10 @@ library_name: transformers
 pipeline_tag: text-generation
 ---
 
-# Wizard Mega 13B
+# Wizard Mega 13B - Pre-Release (Epoch One)
 
 Wizard Mega is a Llama 13B model fine-tuned on the ShareGPT, WizardLM, and Wizard-Vicuna datasets. These particular datasets have all been filtered to remove responses where the model responds with "As an AI language model...", etc or when the model refuses to respond.
+
+# Build
+
+Wizard Mega was built with [Axolotl](https://github.com/OpenAccess-AI-Collective/axolotl) on 8xA100 80GB for 15 hours. The configuration to duplicate this build is provided in this repo's [/config folder](https://huggingface.co/openaccess-ai-collective/wizard-mega-13b/tree/main/configs).

configs/wizard-mega-13b.yml

@@ -0,0 +1,66 @@
+base_model: huggyllama/llama-13b
+base_model_config: huggyllama/llama-13b
+model_type: LlamaForCausalLM
+tokenizer_type: LlamaTokenizer
+load_in_8bit: false
+datasets:
+  - path: anon8231489123/ShareGPT_Vicuna_unfiltered
+    data_files: ShareGPT_V3_unfiltered_cleaned_split_no_imsorry.json
+    type: sharegpt
+  - path: ehartford/WizardLM_alpaca_evol_instruct_70k_unfiltered
+    type: alpaca
+  - path: ehartford/wizard_vicuna_70k_unfiltered
+    type: sharegpt
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.02
+adapter:
+lora_model_dir:
+sequence_len: 2048
+max_packed_sequence_len: 2048
+lora_r: 8
+lora_alpha: 16
+lora_dropout: 0.05
+lora_target_modules:
+  - q_proj
+  - v_proj
+lora_fan_in_fan_out: false
+wandb_project: wizard-mega-13b
+wandb_watch:
+wandb_run_id:
+wandb_log_model:
+output_dir: ./wizard-mega-13b
+batch_size: 512
+micro_batch_size: 8
+num_epochs: 3
+optimizer:
+torchdistx_path:
+lr_scheduler:
+learning_rate: 0.00006
+train_on_inputs: false
+group_by_length: false
+bf16: true
+tf32: true
+gradient_checkpointing: true
+early_stopping_patience:
+resume_from_checkpoint:
+local_rank:
+logging_steps: 1
+xformers_attention: true
+flash_attention:
+gptq_groupsize:
+gptq_model_v1:
+warmup_steps: 20
+eval_steps: 10
+save_steps:
+debug:
+deepspeed:
+weight_decay: 0
+fsdp:
+  - full_shard
+  - auto_wrap
+fsdp_config:
+  fsdp_transformer_layer_cls_to_wrap: LlamaDecoderLayer
+tokens:
+  bos_token: "<s>"
+  eos_token: "</s>"
+  unk_token: "<unk>"