See axolotl config

axolotl version: 0.6.0

base_model: mistralai/Mistral-7B-v0.1
# optionally might have model_type or tokenizer_type
model_type: MistralForCausalLM
tokenizer_type: LlamaTokenizer
# Automatically upload checkpoint and final model to HF
hub_model_id: AiAF/UFOs-Finetune-V1

load_in_8bit: false
load_in_4bit: false
strict: false

datasets:
  - path: json
    data_files: plain_qa_list.jsonl
    ds_type: json
    type: chat_template
    chat_template: chatml
    field_messages: conversations
    message_field_role: from
    message_field_content: value
    roles:
      user:
        - human
      assistant:
        - gpt
      system:
        - system

dataset_prepared_path:
val_set_size: 0.05
output_dir: ./outputs/UFOs-Finetune-V1/out

sequence_len: 8192
sample_packing: true
pad_to_sequence_len: true
eval_sample_packing: false

max_steps: 100000

wandb_project: "UFO_LLM_Finetune"
wandb_entity:
wandb_watch: "all"
wandb_name: "UFO_LLM_Finetune-V1"
wandb_log_model: "false"

gradient_accumulation_steps: 4
micro_batch_size: 2
num_epochs: 10
optimizer: adamw_bnb_8bit
lr_scheduler: cosine
learning_rate: 0.000005

train_on_inputs: false
group_by_length: false
bf16: auto
fp16:
tf32: false

gradient_checkpointing: true
early_stopping_patience:
resume_from_checkpoint: /workspace/axolotl/outputs/out/checkpoint-18
local_rank:
logging_steps: 1
xformers_attention:
flash_attention: true

warmup_steps: 10
evals_per_epoch: 4
eval_table_size:
eval_max_new_tokens: 128
saves_per_epoch: 1
debug:
deepspeed:
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:

UFOs-Finetune-V1

This model is a fine-tuned version of mistralai/Mistral-7B-v0.1 on the json dataset. It achieves the following results on the evaluation set:

• Loss: 1.3935

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: 5e-06
• train_batch_size: 2
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 8
• optimizer: Use OptimizerNames.ADAMW_BNB with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 10
• training_steps: 50

Training results

Training Loss	Epoch	Step	Validation Loss
1.7686	0.1111	1	1.6895
2.0582	0.3333	3	1.6884
1.9135	0.6667	6	1.6793
1.8261	1.0	9	1.6667
1.8757	1.3333	12	1.6570
1.8754	1.6667	15	1.6501
1.8426	2.0	18	1.6468
2.8515	4.1739	21	1.4353
1.3702	4.6957	24	1.4068
1.2889	5.1739	27	1.3909
1.2635	5.6957	30	1.3870
1.2139	6.1739	33	1.3874
1.1786	6.6957	36	1.3895
1.1458	7.1739	39	1.3921
1.1389	7.6957	42	1.3929
1.1255	8.1739	45	1.3934
1.1589	8.6957	48	1.3935

Framework versions

• Transformers 4.48.3
• Pytorch 2.5.1+cu124
• Datasets 3.2.0
• Tokenizers 0.21.0