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