See axolotl config

axolotl version: 0.4.0

base_model: mistralai/Mixtral-8x7B-Instruct-v0.1
model_type: AutoModelForCausalLM
tokenizer_type: LlamaTokenizer
trust_remote_code: true

load_in_8bit: false
load_in_4bit: true
strict: false
chat_template: inst

datasets:
  - path: ./data/with_function_response/more_functions/function_used_training.jsonl
    type: sharegpt
    conversation: mistral
  - path: ./data/with_function_response/more_functions/function_not_used_training.jsonl
    type: sharegpt
    conversation: mistral    
  - path: ./data/with_function_response/parallel_call/missing_parameter_data_training.jsonl
    type: sharegpt
    conversation: mistral
  - path: ./data/with_function_response/parallel_call/parallel_data_training.jsonl
    type: sharegpt
    conversation: mistral

dataset_prepared_path: last_run_prepared
val_set_size: 0.01
output_dir: ../empower-functions-more-tools-parallel

model_config:
  output_router_logits: true

adapter: qlora
lora_model_dir:

sequence_len: 4096
sample_packing: true
pad_to_sequence_len: true

lora_r: 32
lora_alpha: 64
lora_dropout: 0.05
lora_target_modules:
  - q_proj
  - k_proj
  - v_proj
  - o_proj


wandb_project: empower-functions
wandb_name: empower-functions-more-tools-parallel
wandb_log_model: end
hub_model_id: dyang415/empower-functions-more-tools-parallel


gradient_accumulation_steps: 4
micro_batch_size: 2
num_epochs: 4
optimizer: paged_adamw_8bit
lr_scheduler: cosine
learning_rate: 0.0002

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

gradient_checkpointing: true
logging_steps: 1
flash_attention: true

loss_watchdog_threshold: 5.0
loss_watchdog_patience: 3

warmup_steps: 10

eval_table_size:
eval_max_new_tokens: 256
eval_steps: 0.05
save_steps: 0.1
debug:
weight_decay: 0.0
fsdp:
fsdp_config:

empower-functions-more-tools-parallel

This model is a fine-tuned version of mistralai/Mixtral-8x7B-Instruct-v0.1 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.0865

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

The following bitsandbytes quantization config was used during training:

• quant_method: QuantizationMethod.BITS_AND_BYTES
• load_in_8bit: False
• load_in_4bit: True
• llm_int8_threshold: 6.0
• llm_int8_skip_modules: None
• llm_int8_enable_fp32_cpu_offload: False
• llm_int8_has_fp16_weight: False
• bnb_4bit_quant_type: nf4
• bnb_4bit_use_double_quant: True
• bnb_4bit_compute_dtype: bfloat16

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0002
• train_batch_size: 2
• eval_batch_size: 2
• seed: 42
• distributed_type: multi-GPU
• num_devices: 2
• gradient_accumulation_steps: 4
• total_train_batch_size: 16
• total_eval_batch_size: 4
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 10
• num_epochs: 4

Training results

Training Loss	Epoch	Step	Validation Loss
2.0913	0.0	1	2.0864
0.0992	0.2	178	0.1038
0.0923	0.4	356	0.0957
0.0847	0.6	534	0.0938
0.1034	0.8	712	0.0925
0.1062	1.0	890	0.0901
0.1006	1.19	1068	0.0894
0.084	1.39	1246	0.0882
0.0798	1.59	1424	0.0875
0.0752	1.79	1602	0.0849
0.0772	1.99	1780	0.0846
0.0824	2.17	1958	0.0849
0.0792	2.37	2136	0.0843
0.0627	2.57	2314	0.0837
0.0777	2.77	2492	0.0831
0.0636	2.98	2670	0.0827
0.0624	3.16	2848	0.0855
0.0612	3.36	3026	0.0861
0.0649	3.56	3204	0.0861
0.0641	3.76	3382	0.0865

Framework versions

• PEFT 0.7.0
• Transformers 4.37.0
• Pytorch 2.0.1+cu117
• Datasets 2.17.1
• Tokenizers 0.15.0