Support me on Ko-Fi: https://ko-fi.com/deltavector

A series of Merges made for Roleplaying & Creative Writing, This model is a RL train ontop of Archaeo. A merge using Hamanasu-Magnum & Kunou, Trained with Axolotl on 8xH200s.

ChatML formatting

"""<|im_start|>system
system prompt<|im_end|>
<|im_start|>user
Hi there!<|im_end|>
<|im_start|>assistant
Nice to meet you!<|im_end|>
<|im_start|>user
Can I ask a question?<|im_end|>
<|im_start|>assistant
"""

Axolotl Configuration

base_model: ./model

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: true
cut_cross_entropy: false

load_in_8bit: false
load_in_4bit: false
strict: false

rl: kto
kto_undesirable_weight: 1.0

datasets:
  - path: Delta-Vector/Tauri-Opus-Accepted-GPT-Rejected-Opus-Writing-Prompts
    split: train
    type: chatml.argilla
  - path: Delta-Vector/Tauri-IFeval-Dans-Tulu-KTO
    split: train
    type: chatml.argilla
  - path: Delta-Vector/Tauri-KTO-Instruct-Mix
    split: train
    type: chatml.argilla
  - path: Delta-Vector/Tauri-Purpura-Arkhaios-CC-KTO
    split: train
    type: chatml.argilla
dataset_prepared_path: last_run_prepared
val_set_size: 0.0
output_dir: ./archaeo-kto-v2
remove_unused_columns: false

#@lora_mlp_kernel: true
#lora_qkv_kernel: true
#lora_o_kernel: true


adapter: lora
lora_model_dir:

sequence_len: 8192
pad_to_sequence_len: false

lora_r: 64
lora_alpha: 32
lora_dropout: 0.0
lora_target_linear: true
lora_fan_in_fan_out:
lora_target_modules:
  - gate_proj
  - down_proj
  - up_proj
  - q_proj
  - v_proj
  - k_proj
  - o_proj

wandb_project: Francois-V2
wandb_entity: 
wandb_watch:
wandb_name: Archaeo-32b-KTO
wandb_log_model:

gradient_accumulation_steps: 4
micro_batch_size: 4
num_epochs: 1
optimizer: paged_ademamix_8bit
lr_scheduler: constant_with_warmup
learning_rate: 5e-6
max_grad_norm: 0.001

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

gradient_checkpointing: true
gradient_checkpointing_kwargs:
  use_reentrant: true
early_stopping_patience:
resume_from_checkpoint:
local_rank:
logging_steps: 1
xformers_attention:
flash_attention: true

warmup_steps: 100
evals_per_epoch: 4
eval_table_size:
eval_max_new_tokens: 128
saves_per_epoch: 1
debug:
deepspeed: ./deepspeed_configs/zero3_bf16.json
weight_decay: 0.0025
fsdp:
fsdp_config:

Quants:

GGUF
https://huggingface.co/mradermacher/Axo-Merge-Archaeo-V2-Lora-GGUF

EXL2
Delta-Vector/Archaeo-32B-KTO-exl2

Credits

Thank you to: Kubernetes-bad, LucyKnada, Kalomaze, Alicat, Intervitens, Samantha Twinkman, Tav, Trappu & The rest of Anthracite