See axolotl config

axolotl version: 0.8.0.dev0

base_model: mistralai/Mistral-7B-Instruct-v0.3
# 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/Pretrained-QLoRA-Codename-75567-V1

load_in_8bit: false
load_in_4bit: true
strict: false

datasets:
  - path: AiAF/Codename-75567-Pretrainin.jsonl
    type: completion

dataset_prepared_path: last_run_prepared
val_set_size: 0.05
output_dir: ./outputs/qlora-out
save_total_limit: 10

adapter: qlora
lora_model_dir:
lora_r: 256
lora_alpha: 64
lora_dropout: 0.05
lora_target_linear: true
lora_fan_in_fan_out:
lora_target_modules:
  - gate_proj
  - down_proj
  - up_proj
  - q_proj
  - v_proj
  - k_proj
  - o_proj

sequence_len: 512
sample_packing: true
pad_to_sequence_len: true
eval_sample_packing: false

wandb_project: "LLM-Pretraining"
wandb_watch: "all"
wandb_name: "QLoRA-Codename-75567-V1"
wandb_log_model: "false"
wandb_run_id: "QLoRA-Codename-75567-V1"

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:
local_rank:
logging_steps: 1
xformers_attention:
flash_attention: true

warmup_steps: 1
evals_per_epoch: 5
eval_table_size:
eval_max_new_tokens: 128
saves_per_epoch: 1
debug:
deepspeed:
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:

Pretrained-QLoRA-Codename-75567-V1

This model is a fine-tuned version of mistralai/Mistral-7B-Instruct-v0.3 on the AiAF/Codename-75567-Pretrainin.jsonl dataset. It achieves the following results on the evaluation set:

• Loss: 1.6938

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: 2
• num_epochs: 10.0

Training results

Training Loss	Epoch	Step	Validation Loss
1.8916	0.3333	1	1.8880
2.017	0.6667	2	1.8847
1.9119	1.0	3	1.8795
1.9716	1.3333	4	1.8711
1.8532	1.6667	5	1.8601
1.9759	2.0	6	1.8488
1.856	2.3333	7	1.8357
1.8404	2.6667	8	1.8241
1.976	3.0	9	1.8131
1.8504	3.3333	10	1.8012
1.8574	3.6667	11	1.7860
1.8194	4.0	12	1.7749
1.8022	4.3333	13	1.7646
1.7632	4.6667	14	1.7525
1.8326	5.0	15	1.7440
1.7696	5.3333	16	1.7325
1.8039	5.6667	17	1.7257
1.7019	6.0	18	1.7164
1.7878	6.3333	19	1.7132
1.718	6.6667	20	1.7093
1.6994	7.0	21	1.7049
1.785	7.3333	22	1.6996
1.6659	7.6667	23	1.6977
1.7241	8.0	24	1.6970
1.7397	8.3333	25	1.6952
1.6894	8.6667	26	1.6934
1.723	9.0	27	1.6932
1.7999	9.3333	28	1.6927
1.6715	9.6667	29	1.6941
1.6696	10.0	30	1.6938

Framework versions

• PEFT 0.14.0
• Transformers 4.49.0
• Pytorch 2.6.0+cu124
• Datasets 3.2.0
• Tokenizers 0.21.0