Afterparty-hf/Finetune-test1

See axolotl config

axolotl version: 0.4.1

base_model: Afterparty-hf/pretrain-0.924
load_in_8bit: false
load_in_4bit: false
strict: false

datasets:
  - path: Afterparty-hf/synthetic-instruct
    type: sharegpt


  - path: Afterparty-hf/train-format-server
    type: sharegpt


  - path: Afterparty-hf/help-channels-formatted
    type: sharegpt


  - path: Afterparty-hf/constt-augmented
    type: sharegpt


  - path: Afterparty-hf/transcripts-train
    type: sharegpt


chat_template: chatml
dataset_prepared_path: ./prepath
hub_model_id: Afterparty-hf/finetune-0.559
wandb_project: ap_publi
hf_use_auth_token: true


output_dir: ./finetune-559-a
resume_from_checkpoint: ./finetune-559/checkpoint-1026
wandb_watch: all
hub_private_repo: true
hub_strategy: all_checkpoints
push_to_hub: false
hf_use_auth_token: true
max_grad_norm: 0.6
sequence_len: 14256
sample_packing: true
pad_to_sequence_len: true
micro_batch_size: 1
gradient_accumulation_steps: 1
num_epochs: 4
learning_rate: 0.000004
optimizer: adamw_bnb_8bit
#optim_args:
 # amsgrad: true
lr_scheduler: cosine
train_on_inputs: false
group_by_length: false
bfloat16: false
#bf16: auto
fp16:
tf32: false
neftune_noise_alpha: 2
gradient_checkpointing: true
gradient_checkpointing_kwargs:
  use_reentrant: true
logging_steps: 1
xformers_attention:
flash_attention: true
#unsloth_lora_mlp: true
#unsloth_lora_qkv: true
#unsloth_lora_o: true
#flash_attn_cross_entropy: true
#flash_attn_rms_norm: true
#flash_attn_fuse_qkv: false
#flash_attn_fuse_mlp: true
warmup_ratio: 0.5
evals_per_step: 0.025
eval_table_size:
saves_per_epoch: 5
debug:
torch_compile: true
rank:
deepspeed: deepspeed_configs/zero2.json
save_safetensors: true
weight_decay: 0.01
special_tokens:
   bos_token: "<s>"
   eos_token: "</s>"
   unk_token: "<unk>"
   pad_token: "</s>"
tokens: # these are delimiters
  - "<|im_start|>"
  - "<|im_end|>"

finetune-0.559

This model is a fine-tuned version of Afterparty-hf/pretrain-0.924 on the None dataset.

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: 4e-06
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• total_train_batch_size: 8
• total_eval_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 310
• num_epochs: 4

Training results

Framework versions

• Transformers 4.41.1
• Pytorch 2.1.2+cu118
• Datasets 2.19.1
• Tokenizers 0.19.1