axolotl version: 0.4.1

base_model: meta-llama/Meta-Llama-3-8B
model_type: LlamaForCausalLM
tokenizer_type: AutoTokenizer

# load_in_4bit: true

chat_template: chatml
datasets:
  - path: /workspace/datasets/dolphin201-sharegpt2.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/SystemChat_filtered_sharegpt.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/SystemChat_multilingual_sharegpt.jsonl
    type: sharegpt
    conversation: chatml
  # - path: /workspace/datasets/SystemChat-2.0-Arabic/SystemChatArabic_sharegpt.jsonl
  #   type: sharegpt
  #   conversation: chatml
  - path: /workspace/datasets/dolphin-coder-translate-sharegpt2.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/dolphin-coder-codegen-sharegpt2.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/m-a-p_Code-Feedback-sharegpt-unfiltered.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/m-a-p_CodeFeedback-Filtered-Instruction-sharegpt-unfiltered.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/not_samantha_norefusals.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/Orca-Math-resort-unfiltered.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/agent_instruct_react_unfiltered.jsonl
    type: sharegpt  
    conversation: chatml
  - path: /workspace/datasets/toolbench_instruct_j1s1_3k_unfiltered.jsonl
    type: sharegpt  
    conversation: chatml
  - path: /workspace/datasets/toolbench_negative_unfiltered.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/toolbench_react_10p_unfiltered.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/toolbench_tflan_cot_30p_unfiltered.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/openhermes200k_unfiltered.jsonl
    type: sharegpt 
    conversation: chatml

dataset_prepared_path: last_run_prepared
val_set_size: 0.01
output_dir: ./llama-3-8b-2.9.3

sequence_len: 8192
sample_packing: false
pad_to_sequence_len: false

# adapter: qlora
# lora_r: 16
# lora_alpha: 32
# lora_dropout: 0.05
# lora_target_modules:
#   - q_proj
#   - k_proj
#   - v_proj
#   - o_proj
#   - gate_proj
#   - up_proj
#   - down_proj

wandb_project: 2.9.3-llama-3-8b
# wandb_entity: oaaic
# wandb_watch:
# wandb_name:
# wandb_log_model:

gradient_accumulation_steps: 8
micro_batch_size: 2
num_epochs: 3
optimizer: adamw_8bit
lr_scheduler: cosine
learning_rate: 1e-5
# max_grad_norm: 1.0

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

gradient_checkpointing: true
gradient_checkpointing_kwargs:
  use_reentrant: true
logging_steps: 1
flash_attention: true
deepspeed: /workspace/axolotl/deepspeed_configs/zero3_bf16.json
warmup_steps: 10
evals_per_epoch: 2
saves_per_epoch: 2
save_total_limit: 2
weight_decay: 0.1
special_tokens:
  eos_token: "<|im_end|>"
  pad_token: "<|end_of_text|>"
tokens:
  - "<|im_start|>"
  - "<|im_end|>"

llama-3-8b-2.9.3

This model is a fine-tuned version of meta-llama/Meta-Llama-3-8B on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.5771

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-05
• train_batch_size: 2
• eval_batch_size: 2
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 8
• total_train_batch_size: 128
• total_eval_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 10
• num_epochs: 3

Training results

Framework versions

• Transformers 4.42.0.dev0
• Pytorch 2.1.2+cu121
• Datasets 2.19.1
• Tokenizers 0.19.1