See axolotl config

axolotl version: 0.4.0

base_model: KolaGang/v3_pretrain
model_type: MistralForCausalLM
tokenizer_type: LlamaTokenizer
load_in_8bit: false
load_in_4bit: false
strict: false
hub_model_id: KolaGang/KingKan_SFT
hub_strategy: end
datasets:
  - path: KolaGang/Reflection
    type: reflection
  - path: KolaGang/RAG_EAI
    type: context_qa.load_v2
  - path: KolaGang/QA
    type: alpaca_chat.load_qa
  - path: KolaGang/chatlaw
    type: sharegpt
  - path: KolaGang/draft
    type: alpaca
  - path: KolaGang/alpca_w_system
    type: alpaca
  - path: teknium/dataforge-economics
    type: sharegpt
  - path: QuietImpostor/Claude-3-Opus-Claude-3.5-Sonnnet-9k
    type: sharegpt
  - path: Open-Orca/slimorca-deduped-cleaned-corrected
    type: sharegpt
dataset_prepared_path:
val_set_size: 0.05
output_dir: ./outputs/train

sequence_len: 8196
sample_packing: true
pad_to_sequence_len: true
eval_sample_packing: false

wandb_project: mistral_v3
wandb_entity:
wandb_watch:
wandb_name:
wandb_log_model:

gradient_accumulation_steps: 1
micro_batch_size: 4
num_epochs: 3
optimizer: adamw_bnb_8bit
lr_scheduler: cosine
learning_rate: 0.000005

train_on_inputs: false
group_by_length: false
bf16: auto
fp16:
tf32: false

gradient_checkpointing: true
early_stopping_patience:
resume_from_checkpoint:
local_rank:
logging_steps: 1
xformers_attention:
flash_attention: true
flash_attn_cross_entropy: false
flash_attn_rms_norm: true
flash_attn_fuse_qkv: false
flash_attn_fuse_mlp: true

warmup_steps: 10
evals_per_epoch: 4
eval_table_size:
eval_max_new_tokens: 128
saves_per_epoch: 1
debug:
deepspeed: deepspeed_configs/zero2.json
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:
save_safetensors: True

KingKan_SFT

This model is a fine-tuned version of KolaGang/v3_pretrain on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.8390

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-06
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• total_train_batch_size: 32
• total_eval_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 10
• num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss
1.2753	0.0031	1	1.2464
0.8782	0.25	81	0.8252
0.8702	0.5	162	0.8030
0.8288	0.75	243	0.7896
0.8822	1.0	324	0.7797
0.6535	1.2315	405	0.7952
0.6072	1.4815	486	0.7947
0.6683	1.7315	567	0.7914
0.6576	1.9815	648	0.7861
0.4993	2.2130	729	0.8388
0.5151	2.4630	810	0.8383
0.5337	2.7130	891	0.8386
0.4873	2.9630	972	0.8390

Framework versions

• Transformers 4.40.2
• Pytorch 2.1.2+cu118
• Datasets 2.19.1
• Tokenizers 0.19.1