NewEden/Hamanasu-4B-Adventure-CKPTS

See axolotl config

axolotl version: 0.8.0.dev0

base_model: NewEden/Hamanasu-KTO-V2
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer

hub_model_id: NewEden/Hamanasu-4B-Adventure-V2
hub_strategy: "end"
push_dataset_to_hub:
hf_use_auth_token: true

plugins:
  - axolotl.integrations.liger.LigerPlugin
  - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
liger_rope: true
liger_rms_norm: true
liger_layer_norm: true
liger_glu_activation: true
liger_fused_linear_cross_entropy: false
cut_cross_entropy: true


load_in_8bit: false
load_in_4bit: false
strict: false

datasets:
  - path: ./t.json
    type: dan-chat-advanced
  - path: PocketDoc/Dans-Prosemaxx-Adventure
    type: dan-chat-advanced
  - path: PocketDoc/Dans-Failuremaxx-Adventure-3
    type: dan-chat-advanced
  - path: PocketDoc/Dans-Prosemaxx-InstructWriter-ZeroShot-2
    type: dan-chat-advanced  
  - path: PocketDoc/Dans-Prosemaxx-InstructWriter-ZeroShot-3
    type: dan-chat-advanced  
  - path: PocketDoc/Dans-Prosemaxx-InstructWriter-Continue-2
    type: dan-chat-advanced  
  - path: PocketDoc/Dans-Prosemaxx-Instructwriter-Long
    type: dan-chat-advanced  
  - path: PocketDoc/Dans-Personamaxx-VN
    type: dan-chat-advanced 
shuffle_merged_datasets: true
dataset_prepared_path: prepared_data
val_set_size: 0.01
output_dir: ./adventure-2

sequence_len: 32768
sample_packing: true
pad_to_sequence_len: true
eval_sample_packing: False

adapter: lora
lora_model_dir:
lora_r: 128
lora_alpha: 16
lora_dropout: 0.05 
peft_use_rslora: true
lora_target_modules:
  - gate_proj
  - down_proj
  - up_proj
  - q_proj
  - v_proj
  - k_proj
  - o_proj


wandb_project: tavbussy
wandb_entity:
wandb_watch:
wandb_name: adventure-attempt-v2
wandb_log_model:

gradient_accumulation_steps: 2
micro_batch_size: 4
num_epochs: 6
optimizer: adamw_bnb_8bit
lr_scheduler: cosine
learning_rate: 3e-5
max_grad_norm: 0.2

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

warmup_steps: 35
evals_per_epoch: 4
eval_table_size:
eval_max_new_tokens: 128
saves_per_epoch: 2
debug:
deepspeed: ./deepspeed_configs/zero3_bf16.json
weight_decay: 0.03
fsdp:
fsdp_config:
special_tokens:
  pad_token: <|finetune_right_pad_id|>

Hamanasu-4B-Adventure-V2

This model is a fine-tuned version of NewEden/Hamanasu-KTO-V2 on the ./t.json, the PocketDoc/Dans-Prosemaxx-Adventure, the PocketDoc/Dans-Failuremaxx-Adventure-3, the PocketDoc/Dans-Prosemaxx-InstructWriter-ZeroShot-2, the PocketDoc/Dans-Prosemaxx-InstructWriter-ZeroShot-3, the PocketDoc/Dans-Prosemaxx-InstructWriter-Continue-2, the PocketDoc/Dans-Prosemaxx-Instructwriter-Long and the PocketDoc/Dans-Personamaxx-VN datasets. It achieves the following results on the evaluation set:

• Loss: 2.3770

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: 3e-05
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• total_eval_batch_size: 16
• optimizer: Use OptimizerNames.ADAMW_BNB with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 35
• num_epochs: 6.0

Training results

Training Loss	Epoch	Step	Validation Loss
2.5443	0.0083	1	2.5896
2.3222	0.25	30	2.4448
2.3768	0.5	60	2.4210
2.3336	0.75	90	2.4106
2.3595	1.0	120	2.4039
2.3078	1.25	150	2.3991
2.3111	1.5	180	2.3952
2.3331	1.75	210	2.3916
2.2758	2.0	240	2.3889
2.3227	2.25	270	2.3869
2.335	2.5	300	2.3851
2.349	2.75	330	2.3835
2.2828	3.0	360	2.3818
2.2632	3.25	390	2.3811
2.3254	3.5	420	2.3801
2.3073	3.75	450	2.3791
2.337	4.0	480	2.3783
2.2319	4.25	510	2.3782
2.3044	4.5	540	2.3778
2.2948	4.75	570	2.3773
2.2625	5.0	600	2.3771
2.3194	5.25	630	2.3771
2.2457	5.5	660	2.3771
2.2746	5.75	690	2.3769
2.2496	6.0	720	2.3770

Framework versions

• PEFT 0.15.0
• Transformers 4.50.0
• Pytorch 2.5.1+cu124
• Datasets 3.5.0
• Tokenizers 0.21.1