Pretrained-QLoRA-r9kilo-V1

<!-- This model card has been generated automatically according to the information the Trainer had access to. You should probably proofread and complete it, then remove this comment. -->

<img src="https://raw.githubusercontent.com/axolotl-ai-cloud/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>

<details><summary>See axolotl config</summary>

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-r9kilo-V1

load_in_8bit: false
load_in_4bit: true
strict: false

datasets:
  - path: json
    data_files: ["pretraining.jsonl"]
    type: completion

dataset_prepared_path: last_run_prepared
val_set_size: 0.05
output_dir: ./outputs/qlora-out
save_total_limit: 1000

adapter: qlora
lora_model_dir:
lora_r: 256
lora_alpha: 512
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-9000-LLM_Datasets-V1"
wandb_log_model: "false"
wandb_run_id: "QLoRA-9000-LLM_Datasets-V1"

gradient_accumulation_steps: 4
micro_batch_size: 64
num_epochs: 4
optimizer: adamw_bnb_8bit
lr_scheduler: cosine
learning_rate: 0.00000047 #0.0000033 #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: 15
evals_per_epoch: 10
eval_table_size:
eval_max_new_tokens: 128
saves_per_epoch: 5
debug:
deepspeed:
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:

</details><br>

Pretrained-QLoRA-r9kilo-V1

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

• Loss: 2.0777

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: 4.7e-07
• train_batch_size: 64
• eval_batch_size: 64
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 256
• 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: 15
• num_epochs: 4.0

Training results

Framework versions

• PEFT 0.14.0
• Transformers 4.49.0
• Pytorch 2.5.1+cu124
• Datasets 3.2.0
• Tokenizers 0.21.0