wave-on-discord/llama-3-70b-llc-2

See axolotl config

axolotl version: 0.4.0

base_model: NousResearch/Meta-Llama-3-70B
model_type: LlamaForCausalLM
tokenizer_type: PreTrainedTokenizerFast

#overrides_of_model_config:
#  rope_scaling:
#    type: linear
#    factor: 4

special_tokens:
  pad_token: "<|end_of_text|>"

gptq: false
gptq_disable_exllama: true

load_in_8bit: false
load_in_4bit: true
strict: false

datasets:
  - path: /workspace/axolotl/output.jsonl
    ds_type: json
    type: completion
    data_files:
      - /workspace/axolotl/output.jsonl

output_dir: ./lora-out-l3-10

adapter: qlora
lora_model_dir:

sequence_len: 10240
sample_packing: true
eval_sample_packing: true
pad_to_sequence_len: true

lora_r: 32
lora_alpha: 64
lora_dropout: 0.10
lora_target_linear: true
lora_target_modules:
  - gate_proj
  - down_proj
  - up_proj
  - q_proj
  - v_proj
  - k_proj
  - o_proj
peft_use_dora: true

wandb_project: kalomaze-model
wandb_entity:
wandb_watch:
wandb_name:
wandb_log_model:

gradient_accumulation_steps: 6
micro_batch_size: 1
num_epochs: 4
# optimizer: paged_adamw_8bit
# optimizer: adamw_bnb_8bit
optimizer: adamw_bnb_8bit
lr_scheduler: cosine
learning_rate: 0.000015
cosine_min_lr_ratio: 0.2
max_grad_norm: 1.0

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

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

warmup_steps: 0
saves_per_epoch: 6
save_total_limit: 7
debug:
weight_decay: 0.0
# fsdp:
#   - full_shard
#   - auto_wrap
# fsdp_config:
#   fsdp_limit_all_gathers: true
#   fsdp_sync_module_states: true
#   fsdp_offload_params: false
#   fsdp_use_orig_params: false
#   fsdp_cpu_ram_efficient_loading: false
#   fsdp_auto_wrap_policy: TRANSFORMER_BASED_WRAP
#   fsdp_transformer_layer_cls_to_wrap: LlamaDecoderLayer
#   fsdp_state_dict_type: FULL_STATE_DICT

seed: 246

lora-out-l3-10

This model is a fine-tuned version of NousResearch/Meta-Llama-3-70B 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: 1.5e-05
• train_batch_size: 1
• eval_batch_size: 1
• seed: 246
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 6
• total_train_batch_size: 48
• total_eval_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• num_epochs: 4

Training results

Framework versions

• PEFT 0.10.0
• Transformers 4.40.0.dev0
• Pytorch 2.2.1
• Datasets 2.15.0
• Tokenizers 0.15.0