File size: 4,812 Bytes
ce5e5be ac008be c32e2e0 ac008be 8da61b7 ac008be ce5e5be ac008be c40797e ac008be c40797e ac008be c40797e ac008be c40797e ac008be c40797e ac008be c40797e ac008be c40797e ac008be c40797e ac008be c40797e ac008be c40797e ac008be c32e2e0 ac008be c40797e ac008be c40797e ac008be c40797e ac008be c32e2e0 ac008be |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 |
---
license: apache-2.0
library_name: peft
tags:
- axolotl
- generated_from_trainer
base_model: byroneverson/LLaVA-v1.5-7B-rehome
model-index:
- name: LLaVA-v1.5-7B-rehome-qlora
results: []
---
<!-- 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/OpenAccess-AI-Collective/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>](https://github.com/OpenAccess-AI-Collective/axolotl)
<details><summary>See axolotl config</summary>
axolotl version: `0.3.0`
```yaml
base_model: byroneverson/LLaVA-v1.5-7B-rehome
model_type: AutoModelForCausalLM #MistralForCausalLM
tokenizer_type: AutoTokenizer #LlamaTokenizer
#is_mistral_derived_model: true
#
load_in_8bit: false
load_in_4bit: true
strict: false
#
datasets:
- path: yahma/alpaca-cleaned #byroneverson/shell-cmd-instruct
type: completion #solar_shell_instruct #alpaca
# For `completion` datsets only, uses the provided field instead of `text` column
field: output
dataset_prepared_path: last_run_prepared
val_set_size: 0.02
output_dir: /kaggle/working/qlora-out
#
# Push checkpoints to hub
hub_model_id: byroneverson/LLaVA-v1.5-7B-rehome-shell-qlora
# How to push checkpoints to hub
# https://huggingface.co/docs/transformers/v4.31.0/en/main_classes/trainer#transformers.TrainingArguments.hub_strategy
hub_strategy: checkpoint
# Whether to use hf `use_auth_token` for loading datasets. Useful for fetching private datasets
# Required to be true when used in combination with `push_dataset_to_hub`
hf_use_auth_token: true
#
adapter: qlora
lora_model_dir:
#
sequence_len: 2048
sample_packing: true
eval_sample_packing: false
pad_to_sequence_len: true
# Train only these params when performing full train
#unfrozen_parameters: [
# "\\bmlp\\.up_proj\\b",
# "\\bmlp\\.down_proj\\b",
# "\\bmlp\\.gate_proj\\b",
# ]
# LoRA/QLoRA
lora_r: 160 #128
lora_alpha: 80 #384 # A good ballpark is double lora_r for 2x or half for 0.5x
lora_dropout: 0.025
lora_target_linear: false # Only target specified modules
lora_fan_in_fan_out:
lora_target_modules: [
"up_proj",
"down_proj",
"gate_proj",
]
#
wandb_project: "LLaVA-v1.5-7B-rehome-qlora"
wandb_log_model: "checkpoint"
wandb_entity:
wandb_watch:
wandb_run_id:
#
gradient_accumulation_steps: 16 # 1
micro_batch_size: 1
num_epochs: 0.2
optimizer: paged_lion_8bit #paged_adamw_8bit
lr_scheduler: cosine
cosine_min_lr_ratio: 1.0 # Constant lr if 1.0
learning_rate: 0.0001
#
train_on_inputs: false
group_by_length: false
bf16: false #true
fp16: true
tf32: false
#
gradient_checkpointing: true
early_stopping_patience:
# Resume from a specific checkpoint dir
resume_from_checkpoint: #last-checkpoint
# If resume_from_checkpoint isn't set and you simply want it to start where it left off.
# Be careful with this being turned on between different models.
auto_resume_from_checkpoints: true #false
local_rank:
logging_steps: 1
xformers_attention:
# Whether to use flash attention patch https://github.com/Dao-AILab/flash-attention:
flash_attention: false #true
flash_attn_cross_entropy: # Whether to use flash-attention cross entropy implementation - advanced use only
flash_attn_rms_norm: false # Whether to use flash-attention rms norm implementation - advanced use only
flash_attn_fuse_qkv: # Whether to fuse QKV into a single operation
flash_attn_fuse_mlp: # Whether to fuse part of the MLP into a single operation
#
warmup_steps: 4
eval_steps: 50
eval_table_size:
eval_table_max_new_tokens: 128
save_steps:
debug: true #
deepspeed:
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:
bos_token: "<s>"
eos_token: "</s>"
unk_token: "<unk>"
```
</details><br>
# LLaVA-v1.5-7B-rehome-shell-qlora
This model is a fine-tuned version of [byroneverson/LLaVA-v1.5-7B-rehome](https://huggingface.co/byroneverson/LLaVA-v1.5-7B-rehome) on the None dataset.
## 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: 0.0001
- train_batch_size: 1
- eval_batch_size: 1
- seed: 42
- distributed_type: multi-GPU
- num_devices: 2
- gradient_accumulation_steps: 16
- total_train_batch_size: 32
- total_eval_batch_size: 2
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: cosine
- lr_scheduler_warmup_steps: 4
- num_epochs: 0.2
- mixed_precision_training: Native AMP
### Training results
### Framework versions
- PEFT 0.7.2.dev0
- Transformers 4.37.0.dev0
- Pytorch 2.0.1+cu117
- Datasets 2.16.1
- Tokenizers 0.15.0 |