configs/axolotl_config.yml

base_model: mistralai/Mixtral-8x22B-v0.1
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer
tokenizer_use_fast: true

# load_in_8bit: true
# load_in_4bit: false
# strict: false

datasets:
  - path: /workspace/datasets/dolphin-2.9.2/dolphin201-sharegpt2.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9.2/dolphin-coder-codegen-sharegpt2.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9.2/dolphin-coder-translate-sharegpt2.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9.2/m-a-p_Code-Feedback-sharegpt-unfiltered.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9.2/m-a-p_CodeFeedback-Filtered-Instruction-sharegpt-unfiltered.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9.2/not_samantha_norefusals.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9.2/openhermes200k_unfiltered.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9.2/Orca-Math-resort-unfiltered.jsonl
    type: sharegpt  
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9.2/SystemChat_sharegpt.jsonl
    type: sharegpt  
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9.2/toolbench_instruct_j1s1_3k_unfiltered.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9.2/toolbench_negative_unfiltered.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9.2/toolbench_react_10p_unfiltered.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9.2/toolbench_tflan_cot_30p_unfiltered.jsonl
    type: sharegpt 
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9.2/agent_instruct_react_unfiltered.jsonl
    type: sharegpt
    conversation: chatml
  
chat_template: chatml
dataset_prepared_path: mixtral-8x22b-data
val_set_size: 0.01
output_dir: mixtral-8x2b

sequence_len: 16384
sample_packing: true
pad_to_sequence_len: true

unfrozen_parameters:
- ^lm_head.weight$
- ^model.embed_tokens.weight$
- model.layers.54.block_sparse_moe.experts.0.w1
- model.layers.53.block_sparse_moe.experts.0.w1
- model.layers.55.block_sparse_moe.experts.0.w1
- model.layers.51.block_sparse_moe.experts.0.w1
- model.layers.52.block_sparse_moe.experts.0.w1
- model.layers.50.block_sparse_moe.experts.0.w1
- model.layers.47.block_sparse_moe.experts.0.w1
- model.layers.49.block_sparse_moe.experts.0.w1
- model.layers.48.block_sparse_moe.experts.0.w1
- model.layers.46.block_sparse_moe.experts.0.w1
- model.layers.44.block_sparse_moe.experts.0.w1
- model.layers.45.block_sparse_moe.experts.0.w1
- model.layers.13.block_sparse_moe.experts.0.w1
- model.layers.43.block_sparse_moe.experts.0.w1
# block_sparse_moe.experts.0.w2 layers
- model.layers.36.block_sparse_moe.experts.0.w2
- model.layers.42.block_sparse_moe.experts.0.w2
- model.layers.41.block_sparse_moe.experts.0.w2
- model.layers.37.block_sparse_moe.experts.0.w2
- model.layers.34.block_sparse_moe.experts.0.w2
- model.layers.38.block_sparse_moe.experts.0.w2
- model.layers.47.block_sparse_moe.experts.0.w2
- model.layers.35.block_sparse_moe.experts.0.w2
- model.layers.44.block_sparse_moe.experts.0.w2
- model.layers.32.block_sparse_moe.experts.0.w2
- model.layers.40.block_sparse_moe.experts.0.w2
- model.layers.39.block_sparse_moe.experts.0.w2
- model.layers.45.block_sparse_moe.experts.0.w2
- model.layers.33.block_sparse_moe.experts.0.w2
# block_sparse_moe.experts.0.w3 layers
- model.layers.46.block_sparse_moe.experts.0.w3
- model.layers.47.block_sparse_moe.experts.0.w3
- model.layers.44.block_sparse_moe.experts.0.w3
- model.layers.45.block_sparse_moe.experts.0.w3
- model.layers.43.block_sparse_moe.experts.0.w3
- model.layers.49.block_sparse_moe.experts.0.w3
- model.layers.48.block_sparse_moe.experts.0.w3
- model.layers.42.block_sparse_moe.experts.0.w3
- model.layers.36.block_sparse_moe.experts.0.w3
- model.layers.39.block_sparse_moe.experts.0.w3
- model.layers.41.block_sparse_moe.experts.0.w3
- model.layers.38.block_sparse_moe.experts.0.w3
- model.layers.50.block_sparse_moe.experts.0.w3
- model.layers.37.block_sparse_moe.experts.0.w3
# block_sparse_moe.experts.1.w1 layers
- model.layers.54.block_sparse_moe.experts.1.w1
- model.layers.53.block_sparse_moe.experts.1.w1
- model.layers.52.block_sparse_moe.experts.1.w1
- model.layers.51.block_sparse_moe.experts.1.w1
- model.layers.50.block_sparse_moe.experts.1.w1
- model.layers.48.block_sparse_moe.experts.1.w1
- model.layers.49.block_sparse_moe.experts.1.w1
- model.layers.47.block_sparse_moe.experts.1.w1
- model.layers.46.block_sparse_moe.experts.1.w1
- model.layers.7.block_sparse_moe.experts.1.w1
- model.layers.45.block_sparse_moe.experts.1.w1
- model.layers.42.block_sparse_moe.experts.1.w1
- model.layers.12.block_sparse_moe.experts.1.w1
- model.layers.13.block_sparse_moe.experts.1.w1
# block_sparse_moe.experts.1.w2 layers
- model.layers.46.block_sparse_moe.experts.1.w2
- model.layers.37.block_sparse_moe.experts.1.w2
- model.layers.34.block_sparse_moe.experts.1.w2
- model.layers.45.block_sparse_moe.experts.1.w2
- model.layers.43.block_sparse_moe.experts.1.w2
- model.layers.39.block_sparse_moe.experts.1.w2
- model.layers.38.block_sparse_moe.experts.1.w2
- model.layers.42.block_sparse_moe.experts.1.w2
- model.layers.48.block_sparse_moe.experts.1.w2
- model.layers.36.block_sparse_moe.experts.1.w2
- model.layers.40.block_sparse_moe.experts.1.w2
- model.layers.41.block_sparse_moe.experts.1.w2
- model.layers.44.block_sparse_moe.experts.1.w2
- model.layers.33.block_sparse_moe.experts.1.w2
# block_sparse_moe.experts.1.w3 layers
- model.layers.47.block_sparse_moe.experts.1.w3
- model.layers.46.block_sparse_moe.experts.1.w3
- model.layers.48.block_sparse_moe.experts.1.w3
- model.layers.42.block_sparse_moe.experts.1.w3
- model.layers.45.block_sparse_moe.experts.1.w3
- model.layers.40.block_sparse_moe.experts.1.w3
- model.layers.43.block_sparse_moe.experts.1.w3
- model.layers.38.block_sparse_moe.experts.1.w3
- model.layers.39.block_sparse_moe.experts.1.w3
- model.layers.41.block_sparse_moe.experts.1.w3
- model.layers.34.block_sparse_moe.experts.1.w3
- model.layers.37.block_sparse_moe.experts.1.w3
- model.layers.44.block_sparse_moe.experts.1.w3
- model.layers.49.block_sparse_moe.experts.1.w3
# block_sparse_moe.experts.2.w1 layers
- model.layers.53.block_sparse_moe.experts.2.w1
- model.layers.52.block_sparse_moe.experts.2.w1
- model.layers.51.block_sparse_moe.experts.2.w1
- model.layers.54.block_sparse_moe.experts.2.w1
- model.layers.50.block_sparse_moe.experts.2.w1
- model.layers.49.block_sparse_moe.experts.2.w1
- model.layers.48.block_sparse_moe.experts.2.w1
- model.layers.45.block_sparse_moe.experts.2.w1
- model.layers.46.block_sparse_moe.experts.2.w1
- model.layers.47.block_sparse_moe.experts.2.w1
- model.layers.55.block_sparse_moe.experts.2.w1
- model.layers.17.block_sparse_moe.experts.2.w1
- model.layers.43.block_sparse_moe.experts.2.w1
- model.layers.13.block_sparse_moe.experts.2.w1
# block_sparse_moe.experts.2.w2 layers
- model.layers.44.block_sparse_moe.experts.2.w2
- model.layers.34.block_sparse_moe.experts.2.w2
- model.layers.48.block_sparse_moe.experts.2.w2
- model.layers.33.block_sparse_moe.experts.2.w2
- model.layers.39.block_sparse_moe.experts.2.w2
- model.layers.36.block_sparse_moe.experts.2.w2
- model.layers.40.block_sparse_moe.experts.2.w2
- model.layers.32.block_sparse_moe.experts.2.w2
- model.layers.46.block_sparse_moe.experts.2.w2
- model.layers.43.block_sparse_moe.experts.2.w2
- model.layers.37.block_sparse_moe.experts.2.w2
- model.layers.38.block_sparse_moe.experts.2.w2
- model.layers.47.block_sparse_moe.experts.2.w2
- model.layers.42.block_sparse_moe.experts.2.w2
# block_sparse_moe.experts.2.w3 layers
- model.layers.46.block_sparse_moe.experts.2.w3
- model.layers.48.block_sparse_moe.experts.2.w3
- model.layers.45.block_sparse_moe.experts.2.w3
- model.layers.47.block_sparse_moe.experts.2.w3
- model.layers.43.block_sparse_moe.experts.2.w3
- model.layers.49.block_sparse_moe.experts.2.w3
- model.layers.40.block_sparse_moe.experts.2.w3
- model.layers.44.block_sparse_moe.experts.2.w3
- model.layers.39.block_sparse_moe.experts.2.w3
- model.layers.38.block_sparse_moe.experts.2.w3
- model.layers.41.block_sparse_moe.experts.2.w3
- model.layers.52.block_sparse_moe.experts.2.w3
- model.layers.51.block_sparse_moe.experts.2.w3
- model.layers.50.block_sparse_moe.experts.2.w3
# block_sparse_moe.experts.3.w1 layers
- model.layers.54.block_sparse_moe.experts.3.w1
- model.layers.52.block_sparse_moe.experts.3.w1
- model.layers.53.block_sparse_moe.experts.3.w1
- model.layers.51.block_sparse_moe.experts.3.w1
- model.layers.48.block_sparse_moe.experts.3.w1
- model.layers.50.block_sparse_moe.experts.3.w1
- model.layers.49.block_sparse_moe.experts.3.w1
- model.layers.46.block_sparse_moe.experts.3.w1
- model.layers.55.block_sparse_moe.experts.3.w1
- model.layers.47.block_sparse_moe.experts.3.w1
- model.layers.45.block_sparse_moe.experts.3.w1
- model.layers.44.block_sparse_moe.experts.3.w1
- model.layers.12.block_sparse_moe.experts.3.w1
- model.layers.28.block_sparse_moe.experts.3.w1
# block_sparse_moe.experts.3.w2 layers
- model.layers.38.block_sparse_moe.experts.3.w2
- model.layers.37.block_sparse_moe.experts.3.w2
- model.layers.35.block_sparse_moe.experts.3.w2
- model.layers.47.block_sparse_moe.experts.3.w2
- model.layers.39.block_sparse_moe.experts.3.w2
- model.layers.44.block_sparse_moe.experts.3.w2
- model.layers.41.block_sparse_moe.experts.3.w2
- model.layers.43.block_sparse_moe.experts.3.w2
- model.layers.36.block_sparse_moe.experts.3.w2
- model.layers.34.block_sparse_moe.experts.3.w2
- model.layers.33.block_sparse_moe.experts.3.w2
- model.layers.46.block_sparse_moe.experts.3.w2
- model.layers.32.block_sparse_moe.experts.3.w2
- model.layers.40.block_sparse_moe.experts.3.w2
# block_sparse_moe.experts.3.w3 layers
- model.layers.46.block_sparse_moe.experts.3.w3
- model.layers.48.block_sparse_moe.experts.3.w3
- model.layers.45.block_sparse_moe.experts.3.w3
- model.layers.47.block_sparse_moe.experts.3.w3
- model.layers.44.block_sparse_moe.experts.3.w3
- model.layers.43.block_sparse_moe.experts.3.w3
- model.layers.49.block_sparse_moe.experts.3.w3
- model.layers.41.block_sparse_moe.experts.3.w3
- model.layers.39.block_sparse_moe.experts.3.w3
- model.layers.36.block_sparse_moe.experts.3.w3
- model.layers.37.block_sparse_moe.experts.3.w3
- model.layers.50.block_sparse_moe.experts.3.w3
- model.layers.35.block_sparse_moe.experts.3.w3
- model.layers.42.block_sparse_moe.experts.3.w3
# block_sparse_moe.experts.4.w1 layers
- model.layers.52.block_sparse_moe.experts.4.w1
- model.layers.51.block_sparse_moe.experts.4.w1
- model.layers.50.block_sparse_moe.experts.4.w1
- model.layers.53.block_sparse_moe.experts.4.w1
- model.layers.49.block_sparse_moe.experts.4.w1
- model.layers.54.block_sparse_moe.experts.4.w1
- model.layers.48.block_sparse_moe.experts.4.w1
- model.layers.55.block_sparse_moe.experts.4.w1
- model.layers.47.block_sparse_moe.experts.4.w1
- model.layers.44.block_sparse_moe.experts.4.w1
- model.layers.46.block_sparse_moe.experts.4.w1
- model.layers.45.block_sparse_moe.experts.4.w1
- model.layers.12.block_sparse_moe.experts.4.w1
- model.layers.42.block_sparse_moe.experts.4.w1
# block_sparse_moe.experts.4.w2 layers
- model.layers.42.block_sparse_moe.experts.4.w2
- model.layers.44.block_sparse_moe.experts.4.w2
- model.layers.46.block_sparse_moe.experts.4.w2
- model.layers.38.block_sparse_moe.experts.4.w2
- model.layers.34.block_sparse_moe.experts.4.w2
- model.layers.41.block_sparse_moe.experts.4.w2
- model.layers.45.block_sparse_moe.experts.4.w2
- model.layers.32.block_sparse_moe.experts.4.w2
- model.layers.37.block_sparse_moe.experts.4.w2
- model.layers.48.block_sparse_moe.experts.4.w2
- model.layers.36.block_sparse_moe.experts.4.w2
- model.layers.33.block_sparse_moe.experts.4.w2
- model.layers.40.block_sparse_moe.experts.4.w2
- model.layers.30.block_sparse_moe.experts.4.w2
# block_sparse_moe.experts.4.w3 layers
- model.layers.48.block_sparse_moe.experts.4.w3
- model.layers.44.block_sparse_moe.experts.4.w3
- model.layers.47.block_sparse_moe.experts.4.w3
- model.layers.46.block_sparse_moe.experts.4.w3
- model.layers.45.block_sparse_moe.experts.4.w3
- model.layers.49.block_sparse_moe.experts.4.w3
- model.layers.38.block_sparse_moe.experts.4.w3
- model.layers.40.block_sparse_moe.experts.4.w3
- model.layers.43.block_sparse_moe.experts.4.w3
- model.layers.36.block_sparse_moe.experts.4.w3
- model.layers.42.block_sparse_moe.experts.4.w3
- model.layers.41.block_sparse_moe.experts.4.w3
- model.layers.50.block_sparse_moe.experts.4.w3
- model.layers.37.block_sparse_moe.experts.4.w3
# block_sparse_moe.experts.5.w1 layers
- model.layers.54.block_sparse_moe.experts.5.w1
- model.layers.53.block_sparse_moe.experts.5.w1
- model.layers.52.block_sparse_moe.experts.5.w1
- model.layers.51.block_sparse_moe.experts.5.w1
- model.layers.50.block_sparse_moe.experts.5.w1
- model.layers.48.block_sparse_moe.experts.5.w1
- model.layers.49.block_sparse_moe.experts.5.w1
- model.layers.10.block_sparse_moe.experts.5.w1
- model.layers.47.block_sparse_moe.experts.5.w1
- model.layers.55.block_sparse_moe.experts.5.w1
- model.layers.46.block_sparse_moe.experts.5.w1
- model.layers.12.block_sparse_moe.experts.5.w1
- model.layers.44.block_sparse_moe.experts.5.w1
- model.layers.5.block_sparse_moe.experts.5.w1
# block_sparse_moe.experts.5.w2 layers
- model.layers.39.block_sparse_moe.experts.5.w2
- model.layers.32.block_sparse_moe.experts.5.w2
- model.layers.43.block_sparse_moe.experts.5.w2
- model.layers.41.block_sparse_moe.experts.5.w2
- model.layers.46.block_sparse_moe.experts.5.w2
- model.layers.42.block_sparse_moe.experts.5.w2
- model.layers.38.block_sparse_moe.experts.5.w2
- model.layers.34.block_sparse_moe.experts.5.w2
- model.layers.45.block_sparse_moe.experts.5.w2
- model.layers.47.block_sparse_moe.experts.5.w2
- model.layers.36.block_sparse_moe.experts.5.w2
- model.layers.44.block_sparse_moe.experts.5.w2
- model.layers.33.block_sparse_moe.experts.5.w2
- model.layers.35.block_sparse_moe.experts.5.w2
# block_sparse_moe.experts.5.w3 layers
- model.layers.48.block_sparse_moe.experts.5.w3
- model.layers.46.block_sparse_moe.experts.5.w3
- model.layers.47.block_sparse_moe.experts.5.w3
- model.layers.44.block_sparse_moe.experts.5.w3
- model.layers.38.block_sparse_moe.experts.5.w3
- model.layers.41.block_sparse_moe.experts.5.w3
- model.layers.49.block_sparse_moe.experts.5.w3
- model.layers.42.block_sparse_moe.experts.5.w3
- model.layers.40.block_sparse_moe.experts.5.w3
- model.layers.43.block_sparse_moe.experts.5.w3
- model.layers.36.block_sparse_moe.experts.5.w3
- model.layers.39.block_sparse_moe.experts.5.w3
- model.layers.45.block_sparse_moe.experts.5.w3
- model.layers.37.block_sparse_moe.experts.5.w3
# block_sparse_moe.experts.6.w1 layers
- model.layers.54.block_sparse_moe.experts.6.w1
- model.layers.52.block_sparse_moe.experts.6.w1
- model.layers.51.block_sparse_moe.experts.6.w1
- model.layers.53.block_sparse_moe.experts.6.w1
- model.layers.50.block_sparse_moe.experts.6.w1
- model.layers.48.block_sparse_moe.experts.6.w1
- model.layers.49.block_sparse_moe.experts.6.w1
- model.layers.55.block_sparse_moe.experts.6.w1
- model.layers.45.block_sparse_moe.experts.6.w1
- model.layers.47.block_sparse_moe.experts.6.w1
- model.layers.43.block_sparse_moe.experts.6.w1
- model.layers.46.block_sparse_moe.experts.6.w1
- model.layers.13.block_sparse_moe.experts.6.w1
- model.layers.17.block_sparse_moe.experts.6.w1
# block_sparse_moe.experts.6.w2 layers
- model.layers.36.block_sparse_moe.experts.6.w2
- model.layers.38.block_sparse_moe.experts.6.w2
- model.layers.45.block_sparse_moe.experts.6.w2
- model.layers.48.block_sparse_moe.experts.6.w2
- model.layers.44.block_sparse_moe.experts.6.w2
- model.layers.32.block_sparse_moe.experts.6.w2
- model.layers.42.block_sparse_moe.experts.6.w2
- model.layers.40.block_sparse_moe.experts.6.w2
- model.layers.34.block_sparse_moe.experts.6.w2
- model.layers.46.block_sparse_moe.experts.6.w2
- model.layers.41.block_sparse_moe.experts.6.w2
- model.layers.47.block_sparse_moe.experts.6.w2
- model.layers.35.block_sparse_moe.experts.6.w2
- model.layers.39.block_sparse_moe.experts.6.w2
# block_sparse_moe.experts.6.w3 layers
- model.layers.46.block_sparse_moe.experts.6.w3
- model.layers.45.block_sparse_moe.experts.6.w3
- model.layers.43.block_sparse_moe.experts.6.w3
- model.layers.48.block_sparse_moe.experts.6.w3
- model.layers.47.block_sparse_moe.experts.6.w3
- model.layers.37.block_sparse_moe.experts.6.w3
- model.layers.44.block_sparse_moe.experts.6.w3
- model.layers.40.block_sparse_moe.experts.6.w3
- model.layers.41.block_sparse_moe.experts.6.w3
- model.layers.36.block_sparse_moe.experts.6.w3
- model.layers.42.block_sparse_moe.experts.6.w3
- model.layers.38.block_sparse_moe.experts.6.w3
- model.layers.39.block_sparse_moe.experts.6.w3
- model.layers.35.block_sparse_moe.experts.6.w3
# block_sparse_moe.experts.7.w1 layers
- model.layers.54.block_sparse_moe.experts.7.w1
- model.layers.53.block_sparse_moe.experts.7.w1
- model.layers.52.block_sparse_moe.experts.7.w1
- model.layers.51.block_sparse_moe.experts.7.w1
- model.layers.49.block_sparse_moe.experts.7.w1
- model.layers.47.block_sparse_moe.experts.7.w1
- model.layers.48.block_sparse_moe.experts.7.w1
- model.layers.50.block_sparse_moe.experts.7.w1
- model.layers.13.block_sparse_moe.experts.7.w1
- model.layers.45.block_sparse_moe.experts.7.w1
- model.layers.46.block_sparse_moe.experts.7.w1
- model.layers.44.block_sparse_moe.experts.7.w1
- model.layers.18.block_sparse_moe.experts.7.w1
- model.layers.43.block_sparse_moe.experts.7.w1
# block_sparse_moe.experts.7.w2 layers
- model.layers.34.block_sparse_moe.experts.7.w2
- model.layers.33.block_sparse_moe.experts.7.w2
- model.layers.44.block_sparse_moe.experts.7.w2
- model.layers.46.block_sparse_moe.experts.7.w2
- model.layers.41.block_sparse_moe.experts.7.w2
- model.layers.42.block_sparse_moe.experts.7.w2
- model.layers.37.block_sparse_moe.experts.7.w2
- model.layers.39.block_sparse_moe.experts.7.w2
- model.layers.40.block_sparse_moe.experts.7.w2
- model.layers.43.block_sparse_moe.experts.7.w2
- model.layers.35.block_sparse_moe.experts.7.w2
- model.layers.36.block_sparse_moe.experts.7.w2
- model.layers.48.block_sparse_moe.experts.7.w2
- model.layers.38.block_sparse_moe.experts.7.w2
# block_sparse_moe.experts.7.w3 layers
- model.layers.47.block_sparse_moe.experts.7.w3
- model.layers.46.block_sparse_moe.experts.7.w3
- model.layers.45.block_sparse_moe.experts.7.w3
- model.layers.44.block_sparse_moe.experts.7.w3
- model.layers.40.block_sparse_moe.experts.7.w3
- model.layers.48.block_sparse_moe.experts.7.w3
- model.layers.43.block_sparse_moe.experts.7.w3
- model.layers.41.block_sparse_moe.experts.7.w3
- model.layers.39.block_sparse_moe.experts.7.w3
- model.layers.42.block_sparse_moe.experts.7.w3
- model.layers.49.block_sparse_moe.experts.7.w3
- model.layers.31.block_sparse_moe.experts.7.w3
- model.layers.37.block_sparse_moe.experts.7.w3
- model.layers.35.block_sparse_moe.experts.7.w3
# block_sparse_moe.gate layers
- model.layers.0.block_sparse_moe.gate
- model.layers.1.block_sparse_moe.gate
- model.layers.2.block_sparse_moe.gate
- model.layers.3.block_sparse_moe.gate
- model.layers.4.block_sparse_moe.gate
- model.layers.5.block_sparse_moe.gate
- model.layers.6.block_sparse_moe.gate
- model.layers.7.block_sparse_moe.gate
- model.layers.8.block_sparse_moe.gate
- model.layers.9.block_sparse_moe.gate
- model.layers.10.block_sparse_moe.gate
- model.layers.11.block_sparse_moe.gate
- model.layers.12.block_sparse_moe.gate
- model.layers.13.block_sparse_moe.gate
# self_attn.k_proj layers
- model.layers.46.self_attn.k_proj
- model.layers.48.self_attn.k_proj
- model.layers.45.self_attn.k_proj
- model.layers.39.self_attn.k_proj
- model.layers.44.self_attn.k_proj
- model.layers.47.self_attn.k_proj
- model.layers.51.self_attn.k_proj
- model.layers.36.self_attn.k_proj
- model.layers.35.self_attn.k_proj
- model.layers.41.self_attn.k_proj
- model.layers.42.self_attn.k_proj
- model.layers.38.self_attn.k_proj
- model.layers.43.self_attn.k_proj
- model.layers.34.self_attn.k_proj
# self_attn.o_proj layers
- model.layers.20.self_attn.o_proj
- model.layers.19.self_attn.o_proj
- model.layers.16.self_attn.o_proj
- model.layers.13.self_attn.o_proj
- model.layers.18.self_attn.o_proj
- model.layers.17.self_attn.o_proj
- model.layers.42.self_attn.o_proj
- model.layers.12.self_attn.o_proj
- model.layers.14.self_attn.o_proj
- model.layers.15.self_attn.o_proj
- model.layers.40.self_attn.o_proj
- model.layers.22.self_attn.o_proj
- model.layers.23.self_attn.o_proj
- model.layers.38.self_attn.o_proj
# self_attn.q_proj layers
- model.layers.0.self_attn.q_proj
- model.layers.1.self_attn.q_proj
- model.layers.2.self_attn.q_proj
- model.layers.22.self_attn.q_proj
- model.layers.27.self_attn.q_proj
- model.layers.28.self_attn.q_proj
- model.layers.13.self_attn.q_proj
- model.layers.21.self_attn.q_proj
- model.layers.24.self_attn.q_proj
- model.layers.33.self_attn.q_proj
- model.layers.14.self_attn.q_proj
- model.layers.11.self_attn.q_proj
- model.layers.15.self_attn.q_proj
- model.layers.20.self_attn.q_proj
# self_attn.v_proj layers
- model.layers.32.self_attn.v_proj
- model.layers.34.self_attn.v_proj
- model.layers.35.self_attn.v_proj
- model.layers.38.self_attn.v_proj
- model.layers.41.self_attn.v_proj
- model.layers.46.self_attn.v_proj
- model.layers.22.self_attn.v_proj
- model.layers.29.self_attn.v_proj
- model.layers.36.self_attn.v_proj
- model.layers.45.self_attn.v_proj
- model.layers.31.self_attn.v_proj
- model.layers.5.self_attn.v_proj
- model.layers.44.self_attn.v_proj
- model.layers.8.self_attn.v_proj

# adapter: lora
# lora_model_dir:
# lora_r: 32
# lora_alpha: 16
# lora_dropout: 0.05
# lora_target_linear: true
# lora_fan_in_fan_out:

wandb_project: mixtral-8x22b
wandb_entity:
wandb_watch:
wandb_name:
wandb_log_model:

gradient_accumulation_steps: 8
micro_batch_size: 1
num_epochs: 3
optimizer: adamw_8bit
lr_scheduler: cosine
learning_rate: 1e-5

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

warmup_steps: 10
evals_per_epoch: 2
eval_table_size:
eval_max_new_tokens: 128
saves_per_epoch: 4
save_total_limit: 2
debug:
deepspeed: /workspace/axolotl/deepspeed_configs/zero3_bf16_cpuoffload_params.json
weight_decay: 0.1
fsdp:
fsdp_config:
special_tokens:
  eos_token: "<|im_end|>"
  unk_token: "<unk>" 
  bos_token: "<s>"
tokens:
  - "<|im_start|>"