NewEden/DPE-70b-Ckpts

See axolotl config

axolotl version: 0.11.0.dev0

base_model: meta-llama/Llama-3.1-70B
base_model_ignore_patterns: "*/*"
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer

load_in_8bit: false
load_in_4bit: false
strict: false

datasets:
  - path: merged.json
    type: dan-chat-advanced-llama3
  - path: PocketDoc/Dans-Prosemaxx-RP
    type: dan-chat-advanced-llama3
  - path: PocketDoc/Dans-Personamaxx-Logs-2
    type: dan-chat-advanced-llama3
  - path: PocketDoc/Dans-Personamaxx-VN
    type: dan-chat-advanced-llama3
  - path: PocketDoc/Dans-Kinomaxx-VanillaBackrooms
    type: dan-chat-advanced-llama3
  - path: PocketDoc/Dans-Prosemaxx-Gutenberg
    type: dan-chat-advanced-llama3
  - path: PocketDoc/Dans-Prosemaxx-Cowriter-3-XL
    type: dan-chat-advanced-llama3
  - path: PocketDoc/Dans-Prosemaxx-Adventure
    type: dan-chat-advanced-llama3
  - path: PocketDoc/Dans-Failuremaxx-Adventure-3
    type: dan-chat-advanced-llama3
  - path: PocketDoc/Dans-Prosemaxx-InstructWriter-ZeroShot-2
    type: dan-chat-advanced-llama3
  - path: PocketDoc/Dans-Prosemaxx-InstructWriter-ZeroShot-3
    type: dan-chat-advanced-llama3
  - path: PocketDoc/Dans-Prosemaxx-InstructWriter-Continue-2
    type: dan-chat-advanced-llama3
  - path: PocketDoc/Dans-Prosemaxx-Instructwriter-Long
    type: dan-chat-advanced-llama3
  - path: PocketDoc/Dans-Prosemaxx-RepRemover-1
    type: dan-chat-advanced-llama3
  - path: PocketDoc/Dans-MemoryCore-CoreCurriculum-Small
    type: dan-chat-advanced-llama3
  - path: AquaV/US-Army-Survival-Sharegpt
    type: dan-chat-advanced-llama3
  - path: AquaV/Multi-Environment-Operations-Sharegpt
    type: dan-chat-advanced-llama3
  - path: AquaV/Resistance-Sharegpt
    type: dan-chat-advanced-llama3
  - path: AquaV/Interrogation-Sharegpt
    type: dan-chat-advanced-llama3
  - path: AquaV/Chemical-Biological-Safety-Applications-Sharegpt
    type: dan-chat-advanced-llama3
  - path: AquaV/Energetic-Materials-Sharegpt
    type: dan-chat-advanced-llama3
  - path: PocketDoc/Dans-Mathmaxx
    type: dan-chat-advanced-llama3
  - path: PJMixers/Math-Multiturn-1K-ShareGPT
    type: dan-chat-advanced-llama3
  - path: PocketDoc/Dans-Taskmaxx
    type: dan-chat-advanced-llama3
  - path: PocketDoc/Dans-Taskmaxx-DataPrepper
    type: dan-chat-advanced-llama3
  - path: PocketDoc/Dans-Taskmaxx-ConcurrentQA-Reworked
    type: dan-chat-advanced-llama3
  - path: PocketDoc/Dans-Taskmaxx-TableGPT
    type: dan-chat-advanced-llama3
  - path: PocketDoc/Dans-Taskmaxx-SciRIFF
    type: dan-chat-advanced-llama3
  - path: PocketDoc/Dans-Taskmaxx-Edit
    type: dan-chat-advanced-llama3
  - path: PocketDoc/Dans-Toolmaxx-Agent
    type: dan-chat-advanced-llama3
  - path: PocketDoc/Dans-Toolmaxx-ShellCommands
    type: dan-chat-advanced-llama3
  - path: PocketDoc/Dans-Toolmaxx-Functions-Toolbench
    type: dan-chat-advanced-llama3
  - path: PocketDoc/Dans-Toolmaxx-Functions-ToolACE
    type: dan-chat-advanced-llama3
  - path: PocketDoc/Dans-Toolmaxx-Functions-apigen-subset
    type: dan-chat-advanced-llama3
  - path: PocketDoc/Dans-Assistantmaxx-OpenAssistant2
    type: dan-chat-advanced-llama3
  - path: PocketDoc/Dans-Assistantmaxx-Opus-Merge-2
    type: dan-chat-advanced-llama3
  - path: PocketDoc/Dans-Assistantmaxx-sonnetorca-subset
    type: dan-chat-advanced-llama3
  - path: PocketDoc/Dans-Assistantmaxx-sonnetorca-subset-2
    type: dan-chat-advanced-llama3
  - path: PocketDoc/Dans-Assistantmaxx-Synthia
    type: dan-chat-advanced-llama3
  - path: PocketDoc/Dans-Assistantmaxx-ASL
    type: dan-chat-advanced-llama3
  - path: PocketDoc/Dans-Assistantmaxx-PersonaLLM-Opus
    type: dan-chat-advanced-llama3
  - path: PocketDoc/Dans-Assistantmaxx-LongAlign
    type: dan-chat-advanced-llama3
  - path: PocketDoc/Dans-Assistantmaxx-OpenLeecher-Instruct
    type: dan-chat-advanced-llama3
  - path: PocketDoc/Dans-Assistantmaxx-Tulu3-IF
    type: dan-chat-advanced-llama3
  - path: PocketDoc/Dans-Systemmaxx
    type: dan-chat-advanced-llama3
  - path: PocketDoc/Dans-Logicmaxx-SAT-AP
    type: dan-chat-advanced-llama3
  - path: PJMixers/grimulkan_theory-of-mind-ShareGPT
    type: dan-chat-advanced-llama3
  - path: PJMixers/grimulkan_physical-reasoning-ShareGPT
    type: dan-chat-advanced-llama3
  - path: PocketDoc/Dans-Reasoningmaxx-NaturalReasoning
    type: dan-chat-advanced-llama3
  - path: PocketDoc/Dans-Reasoningmaxx-WebInstruct
    type: dan-chat-advanced-llama3
  - path: PocketDoc/Dans-Reasoningmaxx-GeneralReasoning
    type: dan-chat-advanced-llama3
  - path: Delta-Vector/Orion-LN-V1-ShareGPT
    type: dan-chat-advanced-llama3
  - path: Delta-Vector/Orion-Alpindale-LN-ShareGPT
    type: dan-chat-advanced-llama3
  - path: Delta-Vector/Orion-Shoujo-AI-Filtered-ShareGPT
    type: dan-chat-advanced-llama3
  - path: Delta-Vector/Orion-RP-Guild
    type: dan-chat-advanced-llama3
  - path: Delta-Vector/Orion-OpenCAI-ShareGPT
    type: dan-chat-advanced-llama3
  - path: Delta-Vector/Orion-LIMARP-Complexity
    type: dan-chat-advanced-llama3
shuffle_merged_datasets: true
dataset_prepared_path: /home/ubuntu/dpe-70b-data
val_set_size: 0.0
output_dir: /home/ubuntu/dpe-70b-fft-rev1

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

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

wandb_project: dpe-70b-fft
wandb_entity:
wandb_watch:
wandb_name:
wandb_log_model:

gradient_accumulation_steps: 4
micro_batch_size: 1
num_epochs: 2
optimizer: paged_ademamix_8bit
optim_args: "beta1=0.9,beta2=0.999,beta3=0.999,alpha=5"
lr_scheduler: rex
learning_rate: 1.0e-6
max_grad_norm: 0.001

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
s2_attention:

warmup_ratio: 0.1
evals_per_epoch:
eval_table_size:
eval_max_new_tokens:
saves_per_epoch: 2
debug:
deepspeed: ./deepspeed_configs/zero3_bf16.json
weight_decay: 0.0025
fsdp:
fsdp_config:
special_tokens:
  eos_token: <|eot_id|>
  pad_token: <|finetune_right_pad_id|>

home/ubuntu/dpe-70b-fft-rev1

This model is a fine-tuned version of meta-llama/Llama-3.1-70B on the merged.json, the PocketDoc/Dans-Prosemaxx-RP, the PocketDoc/Dans-Personamaxx-Logs-2, the PocketDoc/Dans-Personamaxx-VN, the PocketDoc/Dans-Kinomaxx-VanillaBackrooms, the PocketDoc/Dans-Prosemaxx-Gutenberg, the PocketDoc/Dans-Prosemaxx-Cowriter-3-XL, 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, the PocketDoc/Dans-Prosemaxx-RepRemover-1, the PocketDoc/Dans-MemoryCore-CoreCurriculum-Small, the AquaV/US-Army-Survival-Sharegpt, the AquaV/Multi-Environment-Operations-Sharegpt, the AquaV/Resistance-Sharegpt, the AquaV/Interrogation-Sharegpt, the AquaV/Chemical-Biological-Safety-Applications-Sharegpt, the AquaV/Energetic-Materials-Sharegpt, the PocketDoc/Dans-Mathmaxx, the PJMixers/Math-Multiturn-1K-ShareGPT, the PocketDoc/Dans-Taskmaxx, the PocketDoc/Dans-Taskmaxx-DataPrepper, the PocketDoc/Dans-Taskmaxx-ConcurrentQA-Reworked, the PocketDoc/Dans-Taskmaxx-TableGPT, the PocketDoc/Dans-Taskmaxx-SciRIFF, the PocketDoc/Dans-Taskmaxx-Edit, the PocketDoc/Dans-Toolmaxx-Agent, the PocketDoc/Dans-Toolmaxx-ShellCommands, the PocketDoc/Dans-Toolmaxx-Functions-Toolbench, the PocketDoc/Dans-Toolmaxx-Functions-ToolACE, the PocketDoc/Dans-Toolmaxx-Functions-apigen-subset, the PocketDoc/Dans-Assistantmaxx-OpenAssistant2, the PocketDoc/Dans-Assistantmaxx-Opus-Merge-2, the PocketDoc/Dans-Assistantmaxx-sonnetorca-subset, the PocketDoc/Dans-Assistantmaxx-sonnetorca-subset-2, the PocketDoc/Dans-Assistantmaxx-Synthia, the PocketDoc/Dans-Assistantmaxx-ASL, the PocketDoc/Dans-Assistantmaxx-PersonaLLM-Opus, the PocketDoc/Dans-Assistantmaxx-LongAlign, the PocketDoc/Dans-Assistantmaxx-OpenLeecher-Instruct, the PocketDoc/Dans-Assistantmaxx-Tulu3-IF, the PocketDoc/Dans-Systemmaxx, the PocketDoc/Dans-Logicmaxx-SAT-AP, the PJMixers/grimulkan_theory-of-mind-ShareGPT, the PJMixers/grimulkan_physical-reasoning-ShareGPT, the PocketDoc/Dans-Reasoningmaxx-NaturalReasoning, the PocketDoc/Dans-Reasoningmaxx-WebInstruct, the PocketDoc/Dans-Reasoningmaxx-GeneralReasoning, the Delta-Vector/Orion-LN-V1-ShareGPT, the Delta-Vector/Orion-Alpindale-LN-ShareGPT, the Delta-Vector/Orion-Shoujo-AI-Filtered-ShareGPT, the Delta-Vector/Orion-RP-Guild, the Delta-Vector/Orion-OpenCAI-ShareGPT and the Delta-Vector/Orion-LIMARP-Complexity 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: 1e-06
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 4
• total_train_batch_size: 32
• total_eval_batch_size: 8
• optimizer: Use paged_ademamix_8bit and the args are: beta1=0.9,beta2=0.999,beta3=0.999,alpha=5
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 284
• training_steps: 2848

Training results

Framework versions

• Transformers 4.52.4
• Pytorch 2.7.1+cu128
• Datasets 3.6.0
• Tokenizers 0.21.2