byroneverson
commited on
Commit
•
ac008be
1
Parent(s):
91d82bf
Model save
Browse files
README.md
CHANGED
@@ -1,3 +1,158 @@
|
|
1 |
---
|
2 |
license: apache-2.0
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
license: apache-2.0
|
3 |
+
library_name: peft
|
4 |
+
tags:
|
5 |
+
- generated_from_trainer
|
6 |
+
base_model: byroneverson/LLaVA-v1.5-7B-rehome
|
7 |
+
model-index:
|
8 |
+
- name: LLaVA-v1.5-7B-rehome-shell-qlora
|
9 |
+
results: []
|
10 |
---
|
11 |
+
|
12 |
+
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
|
13 |
+
should probably proofread and complete it, then remove this comment. -->
|
14 |
+
|
15 |
+
[<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)
|
16 |
+
<details><summary>See axolotl config</summary>
|
17 |
+
|
18 |
+
axolotl version: `0.3.0`
|
19 |
+
```yaml
|
20 |
+
base_model: byroneverson/LLaVA-v1.5-7B-rehome
|
21 |
+
model_type: MistralForCausalLM
|
22 |
+
tokenizer_type: LlamaTokenizer
|
23 |
+
#is_mistral_derived_model: true
|
24 |
+
#
|
25 |
+
load_in_8bit: false
|
26 |
+
load_in_4bit: true
|
27 |
+
strict: false
|
28 |
+
#
|
29 |
+
datasets:
|
30 |
+
- path: byroneverson/shell-cmd-instruct
|
31 |
+
type: solar_shell_instruct #alpaca
|
32 |
+
dataset_prepared_path: last_run_prepared
|
33 |
+
val_set_size: 0.05
|
34 |
+
output_dir: ./qlora-out
|
35 |
+
#
|
36 |
+
# Push checkpoints to hub
|
37 |
+
hub_model_id: byroneverson/LLaVA-v1.5-7B-rehome-shell-qlora
|
38 |
+
# How to push checkpoints to hub
|
39 |
+
# https://huggingface.co/docs/transformers/v4.31.0/en/main_classes/trainer#transformers.TrainingArguments.hub_strategy
|
40 |
+
hub_strategy: checkpoint
|
41 |
+
# Whether to use hf `use_auth_token` for loading datasets. Useful for fetching private datasets
|
42 |
+
# Required to be true when used in combination with `push_dataset_to_hub`
|
43 |
+
hf_use_auth_token: true
|
44 |
+
#
|
45 |
+
adapter: qlora
|
46 |
+
lora_model_dir:
|
47 |
+
#
|
48 |
+
sequence_len: 2048
|
49 |
+
sample_packing: true
|
50 |
+
eval_sample_packing: false
|
51 |
+
pad_to_sequence_len: true
|
52 |
+
#
|
53 |
+
lora_r: 128
|
54 |
+
lora_alpha: 16
|
55 |
+
lora_dropout: 0.05
|
56 |
+
lora_target_linear: true
|
57 |
+
lora_fan_in_fan_out:
|
58 |
+
target_modules: [
|
59 |
+
"up_proj",
|
60 |
+
"down_proj",
|
61 |
+
"gate_proj",
|
62 |
+
]
|
63 |
+
#
|
64 |
+
wandb_project: "LLaVA-v1.5-7B-rehome-qlora"
|
65 |
+
wandb_log_model: "checkpoint"
|
66 |
+
wandb_entity:
|
67 |
+
wandb_watch:
|
68 |
+
wandb_run_id:
|
69 |
+
#
|
70 |
+
gradient_accumulation_steps: 2 # 1
|
71 |
+
micro_batch_size: 1
|
72 |
+
num_epochs: 3
|
73 |
+
optimizer: paged_lion_8bit #paged_adamw_8bit
|
74 |
+
lr_scheduler: cosine
|
75 |
+
learning_rate: 0.0002
|
76 |
+
#
|
77 |
+
train_on_inputs: false
|
78 |
+
group_by_length: false
|
79 |
+
bf16: false #true
|
80 |
+
fp16: true
|
81 |
+
tf32: false
|
82 |
+
#
|
83 |
+
gradient_checkpointing: true
|
84 |
+
early_stopping_patience:
|
85 |
+
# Resume from a specific checkpoint dir
|
86 |
+
resume_from_checkpoint: #last-checkpoint
|
87 |
+
# If resume_from_checkpoint isn't set and you simply want it to start where it left off.
|
88 |
+
# Be careful with this being turned on between different models.
|
89 |
+
auto_resume_from_checkpoints: false
|
90 |
+
local_rank:
|
91 |
+
logging_steps: 1
|
92 |
+
xformers_attention:
|
93 |
+
# Whether to use flash attention patch https://github.com/Dao-AILab/flash-attention:
|
94 |
+
flash_attention: false #true
|
95 |
+
flash_attn_cross_entropy: # Whether to use flash-attention cross entropy implementation - advanced use only
|
96 |
+
flash_attn_rms_norm: false # Whether to use flash-attention rms norm implementation - advanced use only
|
97 |
+
flash_attn_fuse_qkv: # Whether to fuse QKV into a single operation
|
98 |
+
flash_attn_fuse_mlp: # Whether to fuse part of the MLP into a single operation
|
99 |
+
#
|
100 |
+
warmup_steps: 10
|
101 |
+
eval_steps: 0.05
|
102 |
+
eval_table_size:
|
103 |
+
eval_table_max_new_tokens: 128
|
104 |
+
save_steps:
|
105 |
+
debug: true #
|
106 |
+
deepspeed:
|
107 |
+
weight_decay: 0.0
|
108 |
+
fsdp:
|
109 |
+
fsdp_config:
|
110 |
+
special_tokens:
|
111 |
+
bos_token: "<s>"
|
112 |
+
eos_token: "</s>"
|
113 |
+
unk_token: "<unk>"
|
114 |
+
|
115 |
+
```
|
116 |
+
|
117 |
+
</details><br>
|
118 |
+
|
119 |
+
# LLaVA-v1.5-7B-rehome-shell-qlora
|
120 |
+
|
121 |
+
This model is a fine-tuned version of [byroneverson/LLaVA-v1.5-7B-rehome](https://huggingface.co/byroneverson/LLaVA-v1.5-7B-rehome) on an unknown dataset.
|
122 |
+
|
123 |
+
## Model description
|
124 |
+
|
125 |
+
More information needed
|
126 |
+
|
127 |
+
## Intended uses & limitations
|
128 |
+
|
129 |
+
More information needed
|
130 |
+
|
131 |
+
## Training and evaluation data
|
132 |
+
|
133 |
+
More information needed
|
134 |
+
|
135 |
+
## Training procedure
|
136 |
+
|
137 |
+
### Training hyperparameters
|
138 |
+
|
139 |
+
The following hyperparameters were used during training:
|
140 |
+
- learning_rate: 0.0002
|
141 |
+
- train_batch_size: 1
|
142 |
+
- eval_batch_size: 1
|
143 |
+
- seed: 42
|
144 |
+
- gradient_accumulation_steps: 2
|
145 |
+
- total_train_batch_size: 2
|
146 |
+
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
147 |
+
- lr_scheduler_type: cosine
|
148 |
+
- lr_scheduler_warmup_steps: 10
|
149 |
+
- num_epochs: 3
|
150 |
+
- mixed_precision_training: Native AMP
|
151 |
+
|
152 |
+
### Framework versions
|
153 |
+
|
154 |
+
- PEFT 0.7.2.dev0
|
155 |
+
- Transformers 4.37.0.dev0
|
156 |
+
- Pytorch 2.0.1+cu117
|
157 |
+
- Datasets 2.16.1
|
158 |
+
- Tokenizers 0.15.0
|