Edens-Gate/brainrot-qwq-ckpts

See axolotl config

axolotl version: 0.8.0.dev0

base_model: NewEden/32B-inst
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer

hub_model_id: NewEden/32b-rp
hub_strategy: "all_checkpoints"
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: NewEden/RP-logs-V2-Experimental-prefixed
    type: dan-chat-advanced
  - path: NewEden/Creative_Writing-Complexity
    type: dan-chat-advanced
  - path: NewEden/Discord-Filtered
    type: dan-chat-advanced
  - path: NewEden/DeepseekRP-Filtered
    type: dan-chat-advanced
  - path: NewEden/Storium-Prefixed-Clean  
    type: dan-chat-advanced
  - path: NewEden/Basket-Weaving-Filtered
    type: dan-chat-advanced
  - path: NewEden/LIMARP-Complexity
    type: dan-chat-advanced
  - path: NewEden/Misc-Data-Sharegpt-Prefixed
    type: dan-chat-advanced
  - path: NewEden/BlueSky-10K-Complexity
    type: dan-chat-advanced 
  - path: NewEden/OpenCAI-ShareGPT
    type: dan-chat-advanced
  - path: NewEden/Basket-Weaving-Filtered
    type: dan-chat-advanced  
  - path: PocketDoc/Dans-Personamaxx-VN
    type: dan-chat-advanced 
  - path: PocketDoc/Dans-Kinomaxx-VanillaBackrooms
    type: dan-chat-advanced  
dataset_prepared_path: prepared_data
val_set_size: 0.0
output_dir: ./qwq-inst

sequence_len: 32768
sample_packing: true
pad_to_sequence_len: true

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

wandb_project: qwq
wandb_entity:
wandb_watch:
wandb_name: rp-attempt-03
wandb_log_model:

gradient_accumulation_steps: 2
micro_batch_size: 2
num_epochs: 4
optimizer: adamw_bnb_8bit
lr_scheduler: cosine
learning_rate: 2.5e-5
max_grad_norm: 1.0

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

gradient_checkpointing: unsloth
early_stopping_patience:
resume_from_checkpoint:
local_rank:
logging_steps: 1
xformers_attention:
flash_attention: true

warmup_steps: 40
saves_per_epoch: 2
debug:
deepspeed: deepspeed_configs/zero3_bf16.json
weight_decay: 0.02
fsdp:
fsdp_config:
special_tokens:

32b-rp

This model is a fine-tuned version of NewEden/32B-inst on the NewEden/RP-logs-V2-Experimental-prefixed, the NewEden/Creative_Writing-Complexity, the NewEden/Discord-Filtered, the NewEden/DeepseekRP-Filtered, the NewEden/Storium-Prefixed-Clean, the NewEden/Basket-Weaving-Filtered, the NewEden/LIMARP-Complexity, the NewEden/Misc-Data-Sharegpt-Prefixed, the NewEden/BlueSky-10K-Complexity, the NewEden/OpenCAI-ShareGPT, the NewEden/Basket-Weaving-Filtered, the PocketDoc/Dans-Personamaxx-VN and the PocketDoc/Dans-Kinomaxx-VanillaBackrooms datasets.

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: 2.5e-05
• train_batch_size: 2
• eval_batch_size: 2
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• 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: 40
• num_epochs: 4.0

Training results

Framework versions

• Transformers 4.49.0
• Pytorch 2.6.0+cu124
• Datasets 3.2.0
• Tokenizers 0.21.0