xianchaowu
commited on
Commit
·
ed1b13e
1
Parent(s):
8b0c479
upload lazy lora adapter for llama-2 70b hf
Browse files- README.md +157 -0
- adapter_config.json +596 -0
- adapter_model.bin +3 -0
- usage.py +51 -0
README.md
CHANGED
@@ -1,3 +1,160 @@
|
|
1 |
---
|
2 |
license: llama2
|
3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
license: llama2
|
3 |
---
|
4 |
+
|
5 |
+
## Lazy LoRA
|
6 |
+
|
7 |
+
### Benefits
|
8 |
+
|
9 |
+
0. using the updated [Meta's LLaMA-2 models](https://huggingface.co/meta-llama/Llama-2-70b-hf).
|
10 |
+
1. support [4-bit qlora](https://arxiv.org/abs/2305.14314), extreme GPU memory and inference time saving;
|
11 |
+
2. comparable MMLU evaluation dataset results, llama2-70b's 68.9% to our 68.21% (-0.69%).
|
12 |
+
3. This lazy-lora adapter is based on [Meta's LLaMA-2-70b-hf](https://huggingface.co/meta-llama/Llama-2-70b-hf), and using the [oasst1 dataset](https://huggingface.co/datasets/OpenAssistant/oasst1), following [Guanaco](https://huggingface.co/timdettmers/guanaco-65b).
|
13 |
+
|
14 |
+
### Introduction
|
15 |
+
Determine the rank of LoRA layers by the singular values of pretrained weight matrices.
|
16 |
+
Also, combines:
|
17 |
+
1. LoRA: [LORA: LOW-RANK ADAPTATION OF LARGE LANGUAGE MODELS](https://arxiv.org/abs/2106.09685)
|
18 |
+
2. Prefix Tuning: [Prefix-Tuning: Optimizing Continuous Prompts for Generation](https://aclanthology.org/2021.acl-long.3
|
19 |
+
53/), [P-Tuning v2: Prompt Tuning Can Be Comparable to Fine-tuning Universally Across Scales and Tasks](https://arxiv.or
|
20 |
+
g/pdf/2110.07602.pdf)
|
21 |
+
3. Prompt Tuning: [The Power of Scale for Parameter-Efficient Prompt Tuning](https://arxiv.org/abs/2104.08691)
|
22 |
+
4. LLaMA adapter: [LLaMA-Adapter: Efficient Fine-tuning of Language Models with Zero-init Attention] (https://arxiv.org/abs/2303.16199)
|
23 |
+
in one model.
|
24 |
+
|
25 |
+
This allows you to perform LoRA (additional low rank adapters inserted to each linear layer), and prompt learning (additional virtual tokens attached to the input and to the attention layers acting as `past_key_values`)
|
26 |
+
|
27 |
+
## Usage:
|
28 |
+
```python
|
29 |
+
import sys
|
30 |
+
sys.path.insert(1, '/workspace/asr/peft/src')
|
31 |
+
# TODO set this path to the lazy-lora source code path,
|
32 |
+
# or you can install it from source code:
|
33 |
+
# TODO, please install lazylora for usage:
|
34 |
+
# git clone [email protected]:Xianchao-Wu/peft.git
|
35 |
+
# cd peft
|
36 |
+
# python setup.py install
|
37 |
+
|
38 |
+
from transformers import (AutoTokenizer,
|
39 |
+
AutoModelForCausalLM, BitsAndBytesConfig)
|
40 |
+
from peft import PeftModel, PeftConfig
|
41 |
+
import os
|
42 |
+
import torch
|
43 |
+
|
44 |
+
#import ipdb; ipdb.set_trace()
|
45 |
+
cache_dir="/workspace/asr/peft/qlora"
|
46 |
+
# TODO set this cache_dir to the path where you
|
47 |
+
# stored (or, want to store) llama2-70b-hf model
|
48 |
+
|
49 |
+
lazylora_dir=os.getcwd()
|
50 |
+
# the path that contains 'adapter_config.json'
|
51 |
+
# and 'adapter_model.bin'
|
52 |
+
|
53 |
+
config = PeftConfig.from_pretrained(lazylora_dir)
|
54 |
+
|
55 |
+
tokenizer = AutoTokenizer.from_pretrained(
|
56 |
+
config.base_model_name_or_path,
|
57 |
+
cache_dir=cache_dir,
|
58 |
+
use_auth_token=True
|
59 |
+
)
|
60 |
+
|
61 |
+
bnb_config = BitsAndBytesConfig(
|
62 |
+
load_in_4bit=True,
|
63 |
+
bnb_4bit_use_double_quant=True,
|
64 |
+
bnb_4bit_quant_type='nf4',
|
65 |
+
bnb_4bit_compute_dtype=torch.bfloat16
|
66 |
+
)
|
67 |
+
|
68 |
+
model = AutoModelForCausalLM.from_pretrained(
|
69 |
+
config.base_model_name_or_path,
|
70 |
+
quantization_config=bnb_config,
|
71 |
+
device_map="auto",
|
72 |
+
cache_dir=cache_dir,
|
73 |
+
use_auth_token=True
|
74 |
+
)
|
75 |
+
#model.print_trainable_parameters()
|
76 |
+
print(sum(p.numel() for p in model.parameters()))
|
77 |
+
# 34,751,127,552 -> half-size of 70B due to 4-bit loading
|
78 |
+
|
79 |
+
model = PeftModel.from_pretrained(model, lazylora_dir)
|
80 |
+
print('after adding lazy lora parameters:')
|
81 |
+
model.print_trainable_parameters()
|
82 |
+
# trainable params: 0 || all params: 35,579,442,176 || trainable%: 0.0
|
83 |
+
```
|
84 |
+
|
85 |
+
## MMLU result:
|
86 |
+
|
87 |
+
```json
|
88 |
+
{"mmlu_loss": 2.3140328107200987,
|
89 |
+
"mmlu_eval_accuracy_abstract_algebra": 0.36363636363636365,
|
90 |
+
"mmlu_eval_accuracy_high_school_chemistry": 0.5,
|
91 |
+
"mmlu_eval_accuracy_college_physics": 0.45454545454545453,
|
92 |
+
"mmlu_eval_accuracy_international_law": 0.9230769230769231,
|
93 |
+
"mmlu_eval_accuracy_nutrition": 0.696969696969697,
|
94 |
+
"mmlu_eval_accuracy_world_religions": 0.8947368421052632,
|
95 |
+
"mmlu_eval_accuracy_medical_genetics": 1.0,
|
96 |
+
"mmlu_eval_accuracy_high_school_computer_science": 0.6666666666666666,
|
97 |
+
"mmlu_eval_accuracy_anatomy": 0.5,
|
98 |
+
"mmlu_eval_accuracy_sociology": 1.0,
|
99 |
+
"mmlu_eval_accuracy_human_sexuality": 0.5833333333333334,
|
100 |
+
"mmlu_eval_accuracy_high_school_world_history": 0.7307692307692307,
|
101 |
+
"mmlu_eval_accuracy_jurisprudence": 0.7272727272727273,
|
102 |
+
"mmlu_eval_accuracy_high_school_mathematics": 0.2413793103448276,
|
103 |
+
"mmlu_eval_accuracy_college_biology": 0.8125,
|
104 |
+
"mmlu_eval_accuracy_machine_learning": 0.5454545454545454,
|
105 |
+
"mmlu_eval_accuracy_us_foreign_policy": 1.0,
|
106 |
+
"mmlu_eval_accuracy_high_school_microeconomics": 0.7692307692307693,
|
107 |
+
"mmlu_eval_accuracy_high_school_us_history": 1.0,
|
108 |
+
"mmlu_eval_accuracy_security_studies": 0.7777777777777778,
|
109 |
+
"mmlu_eval_accuracy_college_chemistry": 0.25,
|
110 |
+
"mmlu_eval_accuracy_college_computer_science": 0.5454545454545454,
|
111 |
+
"mmlu_eval_accuracy_miscellaneous": 0.7790697674418605,
|
112 |
+
"mmlu_eval_accuracy_professional_accounting": 0.7419354838709677,
|
113 |
+
"mmlu_eval_accuracy_business_ethics": 0.7272727272727273,
|
114 |
+
"mmlu_eval_accuracy_electrical_engineering": 0.5625,
|
115 |
+
"mmlu_eval_accuracy_elementary_mathematics": 0.4878048780487805,
|
116 |
+
"mmlu_eval_accuracy_high_school_biology": 0.71875,
|
117 |
+
"mmlu_eval_accuracy_college_mathematics": 0.45454545454545453,
|
118 |
+
"mmlu_eval_accuracy_high_school_european_history": 0.7777777777777778,
|
119 |
+
"mmlu_eval_accuracy_professional_law": 0.5588235294117647,
|
120 |
+
"mmlu_eval_accuracy_prehistory": 0.8,
|
121 |
+
"mmlu_eval_accuracy_high_school_macroeconomics": 0.7674418604651163,
|
122 |
+
"mmlu_eval_accuracy_formal_logic": 0.42857142857142855,
|
123 |
+
"mmlu_eval_accuracy_philosophy": 0.7941176470588235,
|
124 |
+
"mmlu_eval_accuracy_astronomy": 0.75,
|
125 |
+
"mmlu_eval_accuracy_clinical_knowledge": 0.7586206896551724,
|
126 |
+
"mmlu_eval_accuracy_global_facts": 0.5,
|
127 |
+
"mmlu_eval_accuracy_high_school_government_and_politics": 0.9523809523809523,
|
128 |
+
"mmlu_eval_accuracy_moral_disputes": 0.6842105263157895,
|
129 |
+
"mmlu_eval_accuracy_econometrics": 0.5,
|
130 |
+
"mmlu_eval_accuracy_management": 0.9090909090909091,
|
131 |
+
"mmlu_eval_accuracy_high_school_psychology": 0.9666666666666667,
|
132 |
+
"mmlu_eval_accuracy_high_school_geography": 0.9090909090909091,
|
133 |
+
"mmlu_eval_accuracy_human_aging": 0.6956521739130435,
|
134 |
+
"mmlu_eval_accuracy_logical_fallacies": 0.7222222222222222,
|
135 |
+
"mmlu_eval_accuracy_moral_scenarios": 0.49,
|
136 |
+
"mmlu_eval_accuracy_conceptual_physics": 0.5384615384615384,
|
137 |
+
"mmlu_eval_accuracy_professional_psychology": 0.782608695652174,
|
138 |
+
"mmlu_eval_accuracy_college_medicine": 0.7727272727272727,
|
139 |
+
"mmlu_eval_accuracy_high_school_physics": 0.11764705882352941,
|
140 |
+
"mmlu_eval_accuracy_computer_security": 0.7272727272727273,
|
141 |
+
"mmlu_eval_accuracy_virology": 0.5555555555555556,
|
142 |
+
"mmlu_eval_accuracy_professional_medicine": 0.7741935483870968,
|
143 |
+
"mmlu_eval_accuracy_marketing": 0.96,
|
144 |
+
"mmlu_eval_accuracy_public_relations": 0.6666666666666666,
|
145 |
+
"mmlu_eval_accuracy_high_school_statistics": 0.5652173913043478,
|
146 |
+
"mmlu_eval_accuracy": 0.682100004303323,
|
147 |
+
"epoch": 1.7}
|
148 |
+
```
|
149 |
+
|
150 |
+
## License and intended use
|
151 |
+
|
152 |
+
This lazy-lora adapter is based on [Meta's LLaMA-2-70b-hf](https://huggingface.co/meta-llama/Llama-2-70b-hf), and using the [oasst1 dataset](https://huggingface.co/datasets/OpenAssistant/oasst1), following [Guanaco](https://huggingface.co/timdettmers/guanaco-65b).
|
153 |
+
|
154 |
+
lazy lora adapter weights are available under LLAMA-2 license. Note the use of the lazy lora adapter weights, requires access to the LLaMA model weighs. Lazy lora is based on LLaMA and therefore should be used according to the LLaMA license.
|
155 |
+
|
156 |
+
|
157 |
+
## Risks and Biases
|
158 |
+
|
159 |
+
The model can produce factually incorrect output, and should not be relied on to produce factually accurate information. The model was trained on various public datasets; it is possible that this model could generate lewd, biased, or otherwise offensive outputs.
|
160 |
+
|
adapter_config.json
ADDED
@@ -0,0 +1,596 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"base_model_name_or_path": "meta-llama/Llama-2-70b-hf",
|
3 |
+
"bias": "none",
|
4 |
+
"fan_in_fan_out": false,
|
5 |
+
"inference_mode": true,
|
6 |
+
"init_lazy_lora_weights": true,
|
7 |
+
"is_r_by_svd": true,
|
8 |
+
"is_r_reuse": true,
|
9 |
+
"lazy_lora_alpha": 16.0,
|
10 |
+
"lazy_lora_dropout": 0.05,
|
11 |
+
"lazy_pre_adapter_type": "none",
|
12 |
+
"lazy_pre_lora_alpha": 0.1,
|
13 |
+
"modules_to_save": null,
|
14 |
+
"num_attention_heads": 64,
|
15 |
+
"num_layers": 80,
|
16 |
+
"num_transformer_submodules": 1,
|
17 |
+
"num_virtual_tokens": null,
|
18 |
+
"peft_type": "LAZY_LORA",
|
19 |
+
"prefix_tuning_config": null,
|
20 |
+
"prompt_tuning_config": null,
|
21 |
+
"r": 64,
|
22 |
+
"r_by_module_dict": {
|
23 |
+
"model.layers.0.mlp.down_proj": 44,
|
24 |
+
"model.layers.0.mlp.gate_proj": 30,
|
25 |
+
"model.layers.0.mlp.up_proj": 31,
|
26 |
+
"model.layers.0.self_attn.k_proj": 39,
|
27 |
+
"model.layers.0.self_attn.o_proj": 29,
|
28 |
+
"model.layers.0.self_attn.q_proj": 26,
|
29 |
+
"model.layers.0.self_attn.v_proj": 35,
|
30 |
+
"model.layers.1.mlp.down_proj": 57,
|
31 |
+
"model.layers.1.mlp.gate_proj": 42,
|
32 |
+
"model.layers.1.mlp.up_proj": 46,
|
33 |
+
"model.layers.1.self_attn.k_proj": 62,
|
34 |
+
"model.layers.1.self_attn.o_proj": 36,
|
35 |
+
"model.layers.1.self_attn.q_proj": 41,
|
36 |
+
"model.layers.1.self_attn.v_proj": 34,
|
37 |
+
"model.layers.10.mlp.down_proj": 63,
|
38 |
+
"model.layers.10.mlp.gate_proj": 64,
|
39 |
+
"model.layers.10.mlp.up_proj": 64,
|
40 |
+
"model.layers.10.self_attn.k_proj": 67,
|
41 |
+
"model.layers.10.self_attn.o_proj": 60,
|
42 |
+
"model.layers.10.self_attn.q_proj": 68,
|
43 |
+
"model.layers.10.self_attn.v_proj": 56,
|
44 |
+
"model.layers.11.mlp.down_proj": 63,
|
45 |
+
"model.layers.11.mlp.gate_proj": 64,
|
46 |
+
"model.layers.11.mlp.up_proj": 64,
|
47 |
+
"model.layers.11.self_attn.k_proj": 76,
|
48 |
+
"model.layers.11.self_attn.o_proj": 59,
|
49 |
+
"model.layers.11.self_attn.q_proj": 74,
|
50 |
+
"model.layers.11.self_attn.v_proj": 52,
|
51 |
+
"model.layers.12.mlp.down_proj": 63,
|
52 |
+
"model.layers.12.mlp.gate_proj": 64,
|
53 |
+
"model.layers.12.mlp.up_proj": 64,
|
54 |
+
"model.layers.12.self_attn.k_proj": 81,
|
55 |
+
"model.layers.12.self_attn.o_proj": 58,
|
56 |
+
"model.layers.12.self_attn.q_proj": 77,
|
57 |
+
"model.layers.12.self_attn.v_proj": 53,
|
58 |
+
"model.layers.13.mlp.down_proj": 63,
|
59 |
+
"model.layers.13.mlp.gate_proj": 63,
|
60 |
+
"model.layers.13.mlp.up_proj": 64,
|
61 |
+
"model.layers.13.self_attn.k_proj": 74,
|
62 |
+
"model.layers.13.self_attn.o_proj": 62,
|
63 |
+
"model.layers.13.self_attn.q_proj": 75,
|
64 |
+
"model.layers.13.self_attn.v_proj": 51,
|
65 |
+
"model.layers.14.mlp.down_proj": 63,
|
66 |
+
"model.layers.14.mlp.gate_proj": 63,
|
67 |
+
"model.layers.14.mlp.up_proj": 64,
|
68 |
+
"model.layers.14.self_attn.k_proj": 76,
|
69 |
+
"model.layers.14.self_attn.o_proj": 61,
|
70 |
+
"model.layers.14.self_attn.q_proj": 80,
|
71 |
+
"model.layers.14.self_attn.v_proj": 51,
|
72 |
+
"model.layers.15.mlp.down_proj": 63,
|
73 |
+
"model.layers.15.mlp.gate_proj": 63,
|
74 |
+
"model.layers.15.mlp.up_proj": 64,
|
75 |
+
"model.layers.15.self_attn.k_proj": 68,
|
76 |
+
"model.layers.15.self_attn.o_proj": 64,
|
77 |
+
"model.layers.15.self_attn.q_proj": 70,
|
78 |
+
"model.layers.15.self_attn.v_proj": 56,
|
79 |
+
"model.layers.16.mlp.down_proj": 63,
|
80 |
+
"model.layers.16.mlp.gate_proj": 63,
|
81 |
+
"model.layers.16.mlp.up_proj": 64,
|
82 |
+
"model.layers.16.self_attn.k_proj": 73,
|
83 |
+
"model.layers.16.self_attn.o_proj": 63,
|
84 |
+
"model.layers.16.self_attn.q_proj": 71,
|
85 |
+
"model.layers.16.self_attn.v_proj": 55,
|
86 |
+
"model.layers.17.mlp.down_proj": 64,
|
87 |
+
"model.layers.17.mlp.gate_proj": 63,
|
88 |
+
"model.layers.17.mlp.up_proj": 64,
|
89 |
+
"model.layers.17.self_attn.k_proj": 79,
|
90 |
+
"model.layers.17.self_attn.o_proj": 63,
|
91 |
+
"model.layers.17.self_attn.q_proj": 79,
|
92 |
+
"model.layers.17.self_attn.v_proj": 53,
|
93 |
+
"model.layers.18.mlp.down_proj": 64,
|
94 |
+
"model.layers.18.mlp.gate_proj": 63,
|
95 |
+
"model.layers.18.mlp.up_proj": 64,
|
96 |
+
"model.layers.18.self_attn.k_proj": 71,
|
97 |
+
"model.layers.18.self_attn.o_proj": 63,
|
98 |
+
"model.layers.18.self_attn.q_proj": 72,
|
99 |
+
"model.layers.18.self_attn.v_proj": 58,
|
100 |
+
"model.layers.19.mlp.down_proj": 64,
|
101 |
+
"model.layers.19.mlp.gate_proj": 63,
|
102 |
+
"model.layers.19.mlp.up_proj": 64,
|
103 |
+
"model.layers.19.self_attn.k_proj": 73,
|
104 |
+
"model.layers.19.self_attn.o_proj": 68,
|
105 |
+
"model.layers.19.self_attn.q_proj": 74,
|
106 |
+
"model.layers.19.self_attn.v_proj": 61,
|
107 |
+
"model.layers.2.mlp.down_proj": 61,
|
108 |
+
"model.layers.2.mlp.gate_proj": 51,
|
109 |
+
"model.layers.2.mlp.up_proj": 54,
|
110 |
+
"model.layers.2.self_attn.k_proj": 81,
|
111 |
+
"model.layers.2.self_attn.o_proj": 50,
|
112 |
+
"model.layers.2.self_attn.q_proj": 58,
|
113 |
+
"model.layers.2.self_attn.v_proj": 39,
|
114 |
+
"model.layers.20.mlp.down_proj": 64,
|
115 |
+
"model.layers.20.mlp.gate_proj": 63,
|
116 |
+
"model.layers.20.mlp.up_proj": 64,
|
117 |
+
"model.layers.20.self_attn.k_proj": 74,
|
118 |
+
"model.layers.20.self_attn.o_proj": 63,
|
119 |
+
"model.layers.20.self_attn.q_proj": 73,
|
120 |
+
"model.layers.20.self_attn.v_proj": 57,
|
121 |
+
"model.layers.21.mlp.down_proj": 64,
|
122 |
+
"model.layers.21.mlp.gate_proj": 63,
|
123 |
+
"model.layers.21.mlp.up_proj": 65,
|
124 |
+
"model.layers.21.self_attn.k_proj": 66,
|
125 |
+
"model.layers.21.self_attn.o_proj": 63,
|
126 |
+
"model.layers.21.self_attn.q_proj": 65,
|
127 |
+
"model.layers.21.self_attn.v_proj": 56,
|
128 |
+
"model.layers.22.mlp.down_proj": 64,
|
129 |
+
"model.layers.22.mlp.gate_proj": 64,
|
130 |
+
"model.layers.22.mlp.up_proj": 65,
|
131 |
+
"model.layers.22.self_attn.k_proj": 69,
|
132 |
+
"model.layers.22.self_attn.o_proj": 67,
|
133 |
+
"model.layers.22.self_attn.q_proj": 64,
|
134 |
+
"model.layers.22.self_attn.v_proj": 62,
|
135 |
+
"model.layers.23.mlp.down_proj": 63,
|
136 |
+
"model.layers.23.mlp.gate_proj": 64,
|
137 |
+
"model.layers.23.mlp.up_proj": 65,
|
138 |
+
"model.layers.23.self_attn.k_proj": 74,
|
139 |
+
"model.layers.23.self_attn.o_proj": 65,
|
140 |
+
"model.layers.23.self_attn.q_proj": 71,
|
141 |
+
"model.layers.23.self_attn.v_proj": 61,
|
142 |
+
"model.layers.24.mlp.down_proj": 63,
|
143 |
+
"model.layers.24.mlp.gate_proj": 63,
|
144 |
+
"model.layers.24.mlp.up_proj": 65,
|
145 |
+
"model.layers.24.self_attn.k_proj": 75,
|
146 |
+
"model.layers.24.self_attn.o_proj": 60,
|
147 |
+
"model.layers.24.self_attn.q_proj": 70,
|
148 |
+
"model.layers.24.self_attn.v_proj": 56,
|
149 |
+
"model.layers.25.mlp.down_proj": 64,
|
150 |
+
"model.layers.25.mlp.gate_proj": 62,
|
151 |
+
"model.layers.25.mlp.up_proj": 64,
|
152 |
+
"model.layers.25.self_attn.k_proj": 78,
|
153 |
+
"model.layers.25.self_attn.o_proj": 63,
|
154 |
+
"model.layers.25.self_attn.q_proj": 73,
|
155 |
+
"model.layers.25.self_attn.v_proj": 54,
|
156 |
+
"model.layers.26.mlp.down_proj": 64,
|
157 |
+
"model.layers.26.mlp.gate_proj": 63,
|
158 |
+
"model.layers.26.mlp.up_proj": 65,
|
159 |
+
"model.layers.26.self_attn.k_proj": 76,
|
160 |
+
"model.layers.26.self_attn.o_proj": 66,
|
161 |
+
"model.layers.26.self_attn.q_proj": 75,
|
162 |
+
"model.layers.26.self_attn.v_proj": 58,
|
163 |
+
"model.layers.27.mlp.down_proj": 65,
|
164 |
+
"model.layers.27.mlp.gate_proj": 63,
|
165 |
+
"model.layers.27.mlp.up_proj": 65,
|
166 |
+
"model.layers.27.self_attn.k_proj": 68,
|
167 |
+
"model.layers.27.self_attn.o_proj": 64,
|
168 |
+
"model.layers.27.self_attn.q_proj": 62,
|
169 |
+
"model.layers.27.self_attn.v_proj": 62,
|
170 |
+
"model.layers.28.mlp.down_proj": 65,
|
171 |
+
"model.layers.28.mlp.gate_proj": 63,
|
172 |
+
"model.layers.28.mlp.up_proj": 65,
|
173 |
+
"model.layers.28.self_attn.k_proj": 71,
|
174 |
+
"model.layers.28.self_attn.o_proj": 67,
|
175 |
+
"model.layers.28.self_attn.q_proj": 68,
|
176 |
+
"model.layers.28.self_attn.v_proj": 64,
|
177 |
+
"model.layers.29.mlp.down_proj": 65,
|
178 |
+
"model.layers.29.mlp.gate_proj": 63,
|
179 |
+
"model.layers.29.mlp.up_proj": 65,
|
180 |
+
"model.layers.29.self_attn.k_proj": 68,
|
181 |
+
"model.layers.29.self_attn.o_proj": 67,
|
182 |
+
"model.layers.29.self_attn.q_proj": 66,
|
183 |
+
"model.layers.29.self_attn.v_proj": 63,
|
184 |
+
"model.layers.3.mlp.down_proj": 62,
|
185 |
+
"model.layers.3.mlp.gate_proj": 56,
|
186 |
+
"model.layers.3.mlp.up_proj": 58,
|
187 |
+
"model.layers.3.self_attn.k_proj": 83,
|
188 |
+
"model.layers.3.self_attn.o_proj": 59,
|
189 |
+
"model.layers.3.self_attn.q_proj": 70,
|
190 |
+
"model.layers.3.self_attn.v_proj": 46,
|
191 |
+
"model.layers.30.mlp.down_proj": 64,
|
192 |
+
"model.layers.30.mlp.gate_proj": 64,
|
193 |
+
"model.layers.30.mlp.up_proj": 65,
|
194 |
+
"model.layers.30.self_attn.k_proj": 54,
|
195 |
+
"model.layers.30.self_attn.o_proj": 67,
|
196 |
+
"model.layers.30.self_attn.q_proj": 52,
|
197 |
+
"model.layers.30.self_attn.v_proj": 65,
|
198 |
+
"model.layers.31.mlp.down_proj": 64,
|
199 |
+
"model.layers.31.mlp.gate_proj": 64,
|
200 |
+
"model.layers.31.mlp.up_proj": 65,
|
201 |
+
"model.layers.31.self_attn.k_proj": 68,
|
202 |
+
"model.layers.31.self_attn.o_proj": 65,
|
203 |
+
"model.layers.31.self_attn.q_proj": 67,
|
204 |
+
"model.layers.31.self_attn.v_proj": 62,
|
205 |
+
"model.layers.32.mlp.down_proj": 64,
|
206 |
+
"model.layers.32.mlp.gate_proj": 64,
|
207 |
+
"model.layers.32.mlp.up_proj": 65,
|
208 |
+
"model.layers.32.self_attn.k_proj": 72,
|
209 |
+
"model.layers.32.self_attn.o_proj": 63,
|
210 |
+
"model.layers.32.self_attn.q_proj": 71,
|
211 |
+
"model.layers.32.self_attn.v_proj": 62,
|
212 |
+
"model.layers.33.mlp.down_proj": 64,
|
213 |
+
"model.layers.33.mlp.gate_proj": 63,
|
214 |
+
"model.layers.33.mlp.up_proj": 65,
|
215 |
+
"model.layers.33.self_attn.k_proj": 75,
|
216 |
+
"model.layers.33.self_attn.o_proj": 63,
|
217 |
+
"model.layers.33.self_attn.q_proj": 77,
|
218 |
+
"model.layers.33.self_attn.v_proj": 60,
|
219 |
+
"model.layers.34.mlp.down_proj": 64,
|
220 |
+
"model.layers.34.mlp.gate_proj": 63,
|
221 |
+
"model.layers.34.mlp.up_proj": 65,
|
222 |
+
"model.layers.34.self_attn.k_proj": 68,
|
223 |
+
"model.layers.34.self_attn.o_proj": 62,
|
224 |
+
"model.layers.34.self_attn.q_proj": 64,
|
225 |
+
"model.layers.34.self_attn.v_proj": 61,
|
226 |
+
"model.layers.35.mlp.down_proj": 64,
|
227 |
+
"model.layers.35.mlp.gate_proj": 63,
|
228 |
+
"model.layers.35.mlp.up_proj": 65,
|
229 |
+
"model.layers.35.self_attn.k_proj": 66,
|
230 |
+
"model.layers.35.self_attn.o_proj": 61,
|
231 |
+
"model.layers.35.self_attn.q_proj": 67,
|
232 |
+
"model.layers.35.self_attn.v_proj": 60,
|
233 |
+
"model.layers.36.mlp.down_proj": 65,
|
234 |
+
"model.layers.36.mlp.gate_proj": 63,
|
235 |
+
"model.layers.36.mlp.up_proj": 65,
|
236 |
+
"model.layers.36.self_attn.k_proj": 69,
|
237 |
+
"model.layers.36.self_attn.o_proj": 60,
|
238 |
+
"model.layers.36.self_attn.q_proj": 69,
|
239 |
+
"model.layers.36.self_attn.v_proj": 61,
|
240 |
+
"model.layers.37.mlp.down_proj": 65,
|
241 |
+
"model.layers.37.mlp.gate_proj": 63,
|
242 |
+
"model.layers.37.mlp.up_proj": 65,
|
243 |
+
"model.layers.37.self_attn.k_proj": 73,
|
244 |
+
"model.layers.37.self_attn.o_proj": 62,
|
245 |
+
"model.layers.37.self_attn.q_proj": 75,
|
246 |
+
"model.layers.37.self_attn.v_proj": 61,
|
247 |
+
"model.layers.38.mlp.down_proj": 65,
|
248 |
+
"model.layers.38.mlp.gate_proj": 63,
|
249 |
+
"model.layers.38.mlp.up_proj": 65,
|
250 |
+
"model.layers.38.self_attn.k_proj": 72,
|
251 |
+
"model.layers.38.self_attn.o_proj": 60,
|
252 |
+
"model.layers.38.self_attn.q_proj": 76,
|
253 |
+
"model.layers.38.self_attn.v_proj": 59,
|
254 |
+
"model.layers.39.mlp.down_proj": 65,
|
255 |
+
"model.layers.39.mlp.gate_proj": 63,
|
256 |
+
"model.layers.39.mlp.up_proj": 65,
|
257 |
+
"model.layers.39.self_attn.k_proj": 77,
|
258 |
+
"model.layers.39.self_attn.o_proj": 61,
|
259 |
+
"model.layers.39.self_attn.q_proj": 80,
|
260 |
+
"model.layers.39.self_attn.v_proj": 61,
|
261 |
+
"model.layers.4.mlp.down_proj": 62,
|
262 |
+
"model.layers.4.mlp.gate_proj": 59,
|
263 |
+
"model.layers.4.mlp.up_proj": 60,
|
264 |
+
"model.layers.4.self_attn.k_proj": 74,
|
265 |
+
"model.layers.4.self_attn.o_proj": 55,
|
266 |
+
"model.layers.4.self_attn.q_proj": 71,
|
267 |
+
"model.layers.4.self_attn.v_proj": 44,
|
268 |
+
"model.layers.40.mlp.down_proj": 65,
|
269 |
+
"model.layers.40.mlp.gate_proj": 63,
|
270 |
+
"model.layers.40.mlp.up_proj": 65,
|
271 |
+
"model.layers.40.self_attn.k_proj": 73,
|
272 |
+
"model.layers.40.self_attn.o_proj": 68,
|
273 |
+
"model.layers.40.self_attn.q_proj": 79,
|
274 |
+
"model.layers.40.self_attn.v_proj": 65,
|
275 |
+
"model.layers.41.mlp.down_proj": 66,
|
276 |
+
"model.layers.41.mlp.gate_proj": 63,
|
277 |
+
"model.layers.41.mlp.up_proj": 64,
|
278 |
+
"model.layers.41.self_attn.k_proj": 70,
|
279 |
+
"model.layers.41.self_attn.o_proj": 65,
|
280 |
+
"model.layers.41.self_attn.q_proj": 74,
|
281 |
+
"model.layers.41.self_attn.v_proj": 63,
|
282 |
+
"model.layers.42.mlp.down_proj": 65,
|
283 |
+
"model.layers.42.mlp.gate_proj": 64,
|
284 |
+
"model.layers.42.mlp.up_proj": 64,
|
285 |
+
"model.layers.42.self_attn.k_proj": 68,
|
286 |
+
"model.layers.42.self_attn.o_proj": 65,
|
287 |
+
"model.layers.42.self_attn.q_proj": 74,
|
288 |
+
"model.layers.42.self_attn.v_proj": 65,
|
289 |
+
"model.layers.43.mlp.down_proj": 66,
|
290 |
+
"model.layers.43.mlp.gate_proj": 64,
|
291 |
+
"model.layers.43.mlp.up_proj": 64,
|
292 |
+
"model.layers.43.self_attn.k_proj": 69,
|
293 |
+
"model.layers.43.self_attn.o_proj": 65,
|
294 |
+
"model.layers.43.self_attn.q_proj": 70,
|
295 |
+
"model.layers.43.self_attn.v_proj": 64,
|
296 |
+
"model.layers.44.mlp.down_proj": 65,
|
297 |
+
"model.layers.44.mlp.gate_proj": 65,
|
298 |
+
"model.layers.44.mlp.up_proj": 63,
|
299 |
+
"model.layers.44.self_attn.k_proj": 67,
|
300 |
+
"model.layers.44.self_attn.o_proj": 72,
|
301 |
+
"model.layers.44.self_attn.q_proj": 76,
|
302 |
+
"model.layers.44.self_attn.v_proj": 67,
|
303 |
+
"model.layers.45.mlp.down_proj": 65,
|
304 |
+
"model.layers.45.mlp.gate_proj": 65,
|
305 |
+
"model.layers.45.mlp.up_proj": 63,
|
306 |
+
"model.layers.45.self_attn.k_proj": 64,
|
307 |
+
"model.layers.45.self_attn.o_proj": 70,
|
308 |
+
"model.layers.45.self_attn.q_proj": 66,
|
309 |
+
"model.layers.45.self_attn.v_proj": 70,
|
310 |
+
"model.layers.46.mlp.down_proj": 65,
|
311 |
+
"model.layers.46.mlp.gate_proj": 66,
|
312 |
+
"model.layers.46.mlp.up_proj": 63,
|
313 |
+
"model.layers.46.self_attn.k_proj": 61,
|
314 |
+
"model.layers.46.self_attn.o_proj": 71,
|
315 |
+
"model.layers.46.self_attn.q_proj": 63,
|
316 |
+
"model.layers.46.self_attn.v_proj": 69,
|
317 |
+
"model.layers.47.mlp.down_proj": 65,
|
318 |
+
"model.layers.47.mlp.gate_proj": 66,
|
319 |
+
"model.layers.47.mlp.up_proj": 63,
|
320 |
+
"model.layers.47.self_attn.k_proj": 71,
|
321 |
+
"model.layers.47.self_attn.o_proj": 68,
|
322 |
+
"model.layers.47.self_attn.q_proj": 73,
|
323 |
+
"model.layers.47.self_attn.v_proj": 67,
|
324 |
+
"model.layers.48.mlp.down_proj": 65,
|
325 |
+
"model.layers.48.mlp.gate_proj": 67,
|
326 |
+
"model.layers.48.mlp.up_proj": 63,
|
327 |
+
"model.layers.48.self_attn.k_proj": 54,
|
328 |
+
"model.layers.48.self_attn.o_proj": 71,
|
329 |
+
"model.layers.48.self_attn.q_proj": 58,
|
330 |
+
"model.layers.48.self_attn.v_proj": 71,
|
331 |
+
"model.layers.49.mlp.down_proj": 65,
|
332 |
+
"model.layers.49.mlp.gate_proj": 67,
|
333 |
+
"model.layers.49.mlp.up_proj": 64,
|
334 |
+
"model.layers.49.self_attn.k_proj": 53,
|
335 |
+
"model.layers.49.self_attn.o_proj": 67,
|
336 |
+
"model.layers.49.self_attn.q_proj": 50,
|
337 |
+
"model.layers.49.self_attn.v_proj": 70,
|
338 |
+
"model.layers.5.mlp.down_proj": 62,
|
339 |
+
"model.layers.5.mlp.gate_proj": 60,
|
340 |
+
"model.layers.5.mlp.up_proj": 61,
|
341 |
+
"model.layers.5.self_attn.k_proj": 80,
|
342 |
+
"model.layers.5.self_attn.o_proj": 57,
|
343 |
+
"model.layers.5.self_attn.q_proj": 77,
|
344 |
+
"model.layers.5.self_attn.v_proj": 46,
|
345 |
+
"model.layers.50.mlp.down_proj": 65,
|
346 |
+
"model.layers.50.mlp.gate_proj": 67,
|
347 |
+
"model.layers.50.mlp.up_proj": 64,
|
348 |
+
"model.layers.50.self_attn.k_proj": 50,
|
349 |
+
"model.layers.50.self_attn.o_proj": 70,
|
350 |
+
"model.layers.50.self_attn.q_proj": 53,
|
351 |
+
"model.layers.50.self_attn.v_proj": 71,
|
352 |
+
"model.layers.51.mlp.down_proj": 65,
|
353 |
+
"model.layers.51.mlp.gate_proj": 67,
|
354 |
+
"model.layers.51.mlp.up_proj": 64,
|
355 |
+
"model.layers.51.self_attn.k_proj": 56,
|
356 |
+
"model.layers.51.self_attn.o_proj": 68,
|
357 |
+
"model.layers.51.self_attn.q_proj": 59,
|
358 |
+
"model.layers.51.self_attn.v_proj": 72,
|
359 |
+
"model.layers.52.mlp.down_proj": 65,
|
360 |
+
"model.layers.52.mlp.gate_proj": 67,
|
361 |
+
"model.layers.52.mlp.up_proj": 64,
|
362 |
+
"model.layers.52.self_attn.k_proj": 57,
|
363 |
+
"model.layers.52.self_attn.o_proj": 70,
|
364 |
+
"model.layers.52.self_attn.q_proj": 66,
|
365 |
+
"model.layers.52.self_attn.v_proj": 69,
|
366 |
+
"model.layers.53.mlp.down_proj": 65,
|
367 |
+
"model.layers.53.mlp.gate_proj": 67,
|
368 |
+
"model.layers.53.mlp.up_proj": 64,
|
369 |
+
"model.layers.53.self_attn.k_proj": 46,
|
370 |
+
"model.layers.53.self_attn.o_proj": 70,
|
371 |
+
"model.layers.53.self_attn.q_proj": 49,
|
372 |
+
"model.layers.53.self_attn.v_proj": 74,
|
373 |
+
"model.layers.54.mlp.down_proj": 65,
|
374 |
+
"model.layers.54.mlp.gate_proj": 68,
|
375 |
+
"model.layers.54.mlp.up_proj": 64,
|
376 |
+
"model.layers.54.self_attn.k_proj": 46,
|
377 |
+
"model.layers.54.self_attn.o_proj": 69,
|
378 |
+
"model.layers.54.self_attn.q_proj": 50,
|
379 |
+
"model.layers.54.self_attn.v_proj": 73,
|
380 |
+
"model.layers.55.mlp.down_proj": 65,
|
381 |
+
"model.layers.55.mlp.gate_proj": 68,
|
382 |
+
"model.layers.55.mlp.up_proj": 65,
|
383 |
+
"model.layers.55.self_attn.k_proj": 56,
|
384 |
+
"model.layers.55.self_attn.o_proj": 72,
|
385 |
+
"model.layers.55.self_attn.q_proj": 59,
|
386 |
+
"model.layers.55.self_attn.v_proj": 77,
|
387 |
+
"model.layers.56.mlp.down_proj": 65,
|
388 |
+
"model.layers.56.mlp.gate_proj": 68,
|
389 |
+
"model.layers.56.mlp.up_proj": 65,
|
390 |
+
"model.layers.56.self_attn.k_proj": 55,
|
391 |
+
"model.layers.56.self_attn.o_proj": 68,
|
392 |
+
"model.layers.56.self_attn.q_proj": 56,
|
393 |
+
"model.layers.56.self_attn.v_proj": 73,
|
394 |
+
"model.layers.57.mlp.down_proj": 65,
|
395 |
+
"model.layers.57.mlp.gate_proj": 68,
|
396 |
+
"model.layers.57.mlp.up_proj": 65,
|
397 |
+
"model.layers.57.self_attn.k_proj": 48,
|
398 |
+
"model.layers.57.self_attn.o_proj": 72,
|
399 |
+
"model.layers.57.self_attn.q_proj": 52,
|
400 |
+
"model.layers.57.self_attn.v_proj": 77,
|
401 |
+
"model.layers.58.mlp.down_proj": 65,
|
402 |
+
"model.layers.58.mlp.gate_proj": 68,
|
403 |
+
"model.layers.58.mlp.up_proj": 65,
|
404 |
+
"model.layers.58.self_attn.k_proj": 44,
|
405 |
+
"model.layers.58.self_attn.o_proj": 69,
|
406 |
+
"model.layers.58.self_attn.q_proj": 46,
|
407 |
+
"model.layers.58.self_attn.v_proj": 73,
|
408 |
+
"model.layers.59.mlp.down_proj": 65,
|
409 |
+
"model.layers.59.mlp.gate_proj": 68,
|
410 |
+
"model.layers.59.mlp.up_proj": 65,
|
411 |
+
"model.layers.59.self_attn.k_proj": 45,
|
412 |
+
"model.layers.59.self_attn.o_proj": 68,
|
413 |
+
"model.layers.59.self_attn.q_proj": 47,
|
414 |
+
"model.layers.59.self_attn.v_proj": 74,
|
415 |
+
"model.layers.6.mlp.down_proj": 62,
|
416 |
+
"model.layers.6.mlp.gate_proj": 60,
|
417 |
+
"model.layers.6.mlp.up_proj": 61,
|
418 |
+
"model.layers.6.self_attn.k_proj": 80,
|
419 |
+
"model.layers.6.self_attn.o_proj": 61,
|
420 |
+
"model.layers.6.self_attn.q_proj": 80,
|
421 |
+
"model.layers.6.self_attn.v_proj": 47,
|
422 |
+
"model.layers.60.mlp.down_proj": 65,
|
423 |
+
"model.layers.60.mlp.gate_proj": 68,
|
424 |
+
"model.layers.60.mlp.up_proj": 66,
|
425 |
+
"model.layers.60.self_attn.k_proj": 27,
|
426 |
+
"model.layers.60.self_attn.o_proj": 67,
|
427 |
+
"model.layers.60.self_attn.q_proj": 28,
|
428 |
+
"model.layers.60.self_attn.v_proj": 74,
|
429 |
+
"model.layers.61.mlp.down_proj": 65,
|
430 |
+
"model.layers.61.mlp.gate_proj": 68,
|
431 |
+
"model.layers.61.mlp.up_proj": 66,
|
432 |
+
"model.layers.61.self_attn.k_proj": 41,
|
433 |
+
"model.layers.61.self_attn.o_proj": 70,
|
434 |
+
"model.layers.61.self_attn.q_proj": 43,
|
435 |
+
"model.layers.61.self_attn.v_proj": 76,
|
436 |
+
"model.layers.62.mlp.down_proj": 65,
|
437 |
+
"model.layers.62.mlp.gate_proj": 68,
|
438 |
+
"model.layers.62.mlp.up_proj": 66,
|
439 |
+
"model.layers.62.self_attn.k_proj": 42,
|
440 |
+
"model.layers.62.self_attn.o_proj": 68,
|
441 |
+
"model.layers.62.self_attn.q_proj": 42,
|
442 |
+
"model.layers.62.self_attn.v_proj": 75,
|
443 |
+
"model.layers.63.mlp.down_proj": 65,
|
444 |
+
"model.layers.63.mlp.gate_proj": 68,
|
445 |
+
"model.layers.63.mlp.up_proj": 66,
|
446 |
+
"model.layers.63.self_attn.k_proj": 40,
|
447 |
+
"model.layers.63.self_attn.o_proj": 65,
|
448 |
+
"model.layers.63.self_attn.q_proj": 43,
|
449 |
+
"model.layers.63.self_attn.v_proj": 72,
|
450 |
+
"model.layers.64.mlp.down_proj": 65,
|
451 |
+
"model.layers.64.mlp.gate_proj": 68,
|
452 |
+
"model.layers.64.mlp.up_proj": 66,
|
453 |
+
"model.layers.64.self_attn.k_proj": 50,
|
454 |
+
"model.layers.64.self_attn.o_proj": 71,
|
455 |
+
"model.layers.64.self_attn.q_proj": 51,
|
456 |
+
"model.layers.64.self_attn.v_proj": 77,
|
457 |
+
"model.layers.65.mlp.down_proj": 65,
|
458 |
+
"model.layers.65.mlp.gate_proj": 68,
|
459 |
+
"model.layers.65.mlp.up_proj": 66,
|
460 |
+
"model.layers.65.self_attn.k_proj": 33,
|
461 |
+
"model.layers.65.self_attn.o_proj": 64,
|
462 |
+
"model.layers.65.self_attn.q_proj": 35,
|
463 |
+
"model.layers.65.self_attn.v_proj": 71,
|
464 |
+
"model.layers.66.mlp.down_proj": 65,
|
465 |
+
"model.layers.66.mlp.gate_proj": 68,
|
466 |
+
"model.layers.66.mlp.up_proj": 67,
|
467 |
+
"model.layers.66.self_attn.k_proj": 41,
|
468 |
+
"model.layers.66.self_attn.o_proj": 68,
|
469 |
+
"model.layers.66.self_attn.q_proj": 43,
|
470 |
+
"model.layers.66.self_attn.v_proj": 74,
|
471 |
+
"model.layers.67.mlp.down_proj": 65,
|
472 |
+
"model.layers.67.mlp.gate_proj": 68,
|
473 |
+
"model.layers.67.mlp.up_proj": 67,
|
474 |
+
"model.layers.67.self_attn.k_proj": 35,
|
475 |
+
"model.layers.67.self_attn.o_proj": 57,
|
476 |
+
"model.layers.67.self_attn.q_proj": 34,
|
477 |
+
"model.layers.67.self_attn.v_proj": 67,
|
478 |
+
"model.layers.68.mlp.down_proj": 65,
|
479 |
+
"model.layers.68.mlp.gate_proj": 68,
|
480 |
+
"model.layers.68.mlp.up_proj": 67,
|
481 |
+
"model.layers.68.self_attn.k_proj": 54,
|
482 |
+
"model.layers.68.self_attn.o_proj": 71,
|
483 |
+
"model.layers.68.self_attn.q_proj": 56,
|
484 |
+
"model.layers.68.self_attn.v_proj": 80,
|
485 |
+
"model.layers.69.mlp.down_proj": 65,
|
486 |
+
"model.layers.69.mlp.gate_proj": 68,
|
487 |
+
"model.layers.69.mlp.up_proj": 67,
|
488 |
+
"model.layers.69.self_attn.k_proj": 65,
|
489 |
+
"model.layers.69.self_attn.o_proj": 72,
|
490 |
+
"model.layers.69.self_attn.q_proj": 65,
|
491 |
+
"model.layers.69.self_attn.v_proj": 84,
|
492 |
+
"model.layers.7.mlp.down_proj": 63,
|
493 |
+
"model.layers.7.mlp.gate_proj": 61,
|
494 |
+
"model.layers.7.mlp.up_proj": 62,
|
495 |
+
"model.layers.7.self_attn.k_proj": 81,
|
496 |
+
"model.layers.7.self_attn.o_proj": 59,
|
497 |
+
"model.layers.7.self_attn.q_proj": 82,
|
498 |
+
"model.layers.7.self_attn.v_proj": 45,
|
499 |
+
"model.layers.70.mlp.down_proj": 65,
|
500 |
+
"model.layers.70.mlp.gate_proj": 68,
|
501 |
+
"model.layers.70.mlp.up_proj": 68,
|
502 |
+
"model.layers.70.self_attn.k_proj": 58,
|
503 |
+
"model.layers.70.self_attn.o_proj": 66,
|
504 |
+
"model.layers.70.self_attn.q_proj": 60,
|
505 |
+
"model.layers.70.self_attn.v_proj": 77,
|
506 |
+
"model.layers.71.mlp.down_proj": 65,
|
507 |
+
"model.layers.71.mlp.gate_proj": 68,
|
508 |
+
"model.layers.71.mlp.up_proj": 68,
|
509 |
+
"model.layers.71.self_attn.k_proj": 65,
|
510 |
+
"model.layers.71.self_attn.o_proj": 68,
|
511 |
+
"model.layers.71.self_attn.q_proj": 65,
|
512 |
+
"model.layers.71.self_attn.v_proj": 77,
|
513 |
+
"model.layers.72.mlp.down_proj": 65,
|
514 |
+
"model.layers.72.mlp.gate_proj": 67,
|
515 |
+
"model.layers.72.mlp.up_proj": 68,
|
516 |
+
"model.layers.72.self_attn.k_proj": 70,
|
517 |
+
"model.layers.72.self_attn.o_proj": 69,
|
518 |
+
"model.layers.72.self_attn.q_proj": 72,
|
519 |
+
"model.layers.72.self_attn.v_proj": 77,
|
520 |
+
"model.layers.73.mlp.down_proj": 65,
|
521 |
+
"model.layers.73.mlp.gate_proj": 67,
|
522 |
+
"model.layers.73.mlp.up_proj": 68,
|
523 |
+
"model.layers.73.self_attn.k_proj": 67,
|
524 |
+
"model.layers.73.self_attn.o_proj": 69,
|
525 |
+
"model.layers.73.self_attn.q_proj": 68,
|
526 |
+
"model.layers.73.self_attn.v_proj": 82,
|
527 |
+
"model.layers.74.mlp.down_proj": 65,
|
528 |
+
"model.layers.74.mlp.gate_proj": 67,
|
529 |
+
"model.layers.74.mlp.up_proj": 68,
|
530 |
+
"model.layers.74.self_attn.k_proj": 68,
|
531 |
+
"model.layers.74.self_attn.o_proj": 65,
|
532 |
+
"model.layers.74.self_attn.q_proj": 70,
|
533 |
+
"model.layers.74.self_attn.v_proj": 77,
|
534 |
+
"model.layers.75.mlp.down_proj": 65,
|
535 |
+
"model.layers.75.mlp.gate_proj": 67,
|
536 |
+
"model.layers.75.mlp.up_proj": 68,
|
537 |
+
"model.layers.75.self_attn.k_proj": 64,
|
538 |
+
"model.layers.75.self_attn.o_proj": 65,
|
539 |
+
"model.layers.75.self_attn.q_proj": 68,
|
540 |
+
"model.layers.75.self_attn.v_proj": 82,
|
541 |
+
"model.layers.76.mlp.down_proj": 65,
|
542 |
+
"model.layers.76.mlp.gate_proj": 67,
|
543 |
+
"model.layers.76.mlp.up_proj": 68,
|
544 |
+
"model.layers.76.self_attn.k_proj": 67,
|
545 |
+
"model.layers.76.self_attn.o_proj": 71,
|
546 |
+
"model.layers.76.self_attn.q_proj": 75,
|
547 |
+
"model.layers.76.self_attn.v_proj": 89,
|
548 |
+
"model.layers.77.mlp.down_proj": 65,
|
549 |
+
"model.layers.77.mlp.gate_proj": 67,
|
550 |
+
"model.layers.77.mlp.up_proj": 68,
|
551 |
+
"model.layers.77.self_attn.k_proj": 64,
|
552 |
+
"model.layers.77.self_attn.o_proj": 65,
|
553 |
+
"model.layers.77.self_attn.q_proj": 69,
|
554 |
+
"model.layers.77.self_attn.v_proj": 84,
|
555 |
+
"model.layers.78.mlp.down_proj": 64,
|
556 |
+
"model.layers.78.mlp.gate_proj": 67,
|
557 |
+
"model.layers.78.mlp.up_proj": 68,
|
558 |
+
"model.layers.78.self_attn.k_proj": 64,
|
559 |
+
"model.layers.78.self_attn.o_proj": 60,
|
560 |
+
"model.layers.78.self_attn.q_proj": 66,
|
561 |
+
"model.layers.78.self_attn.v_proj": 77,
|
562 |
+
"model.layers.79.mlp.down_proj": 63,
|
563 |
+
"model.layers.79.mlp.gate_proj": 68,
|
564 |
+
"model.layers.79.mlp.up_proj": 67,
|
565 |
+
"model.layers.79.self_attn.k_proj": 65,
|
566 |
+
"model.layers.79.self_attn.o_proj": 48,
|
567 |
+
"model.layers.79.self_attn.q_proj": 61,
|
568 |
+
"model.layers.79.self_attn.v_proj": 62,
|
569 |
+
"model.layers.8.mlp.down_proj": 64,
|
570 |
+
"model.layers.8.mlp.gate_proj": 62,
|
571 |
+
"model.layers.8.mlp.up_proj": 62,
|
572 |
+
"model.layers.8.self_attn.k_proj": 82,
|
573 |
+
"model.layers.8.self_attn.o_proj": 61,
|
574 |
+
"model.layers.8.self_attn.q_proj": 82,
|
575 |
+
"model.layers.8.self_attn.v_proj": 46,
|
576 |
+
"model.layers.9.mlp.down_proj": 64,
|
577 |
+
"model.layers.9.mlp.gate_proj": 63,
|
578 |
+
"model.layers.9.mlp.up_proj": 63,
|
579 |
+
"model.layers.9.self_attn.k_proj": 77,
|
580 |
+
"model.layers.9.self_attn.o_proj": 62,
|
581 |
+
"model.layers.9.self_attn.q_proj": 76,
|
582 |
+
"model.layers.9.self_attn.v_proj": 51
|
583 |
+
},
|
584 |
+
"rank_file": "/workspace/asr/peft/qlora/llama2_70bhf_lazylora_r64_2.json",
|
585 |
+
"target_modules": [
|
586 |
+
"o_proj",
|
587 |
+
"up_proj",
|
588 |
+
"k_proj",
|
589 |
+
"down_proj",
|
590 |
+
"q_proj",
|
591 |
+
"v_proj",
|
592 |
+
"gate_proj"
|
593 |
+
],
|
594 |
+
"task_type": "CAUSAL_LM",
|
595 |
+
"token_dim": 8192
|
596 |
+
}
|
adapter_model.bin
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:857b3b46b329f38aadcc7ddf874025e17ec34f657ce778cd4a8e4a589455bbce
|
3 |
+
size 1657058821
|
usage.py
ADDED
@@ -0,0 +1,51 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import sys
|
2 |
+
sys.path.insert(1, '/workspace/asr/peft/src')
|
3 |
+
# TODO set this path to the lazy-lora source code path, or you can install it from source code:
|
4 |
+
# TODO, please install lazylora for usage:
|
5 |
+
# git clone [email protected]:Xianchao-Wu/peft.git
|
6 |
+
# cd peft
|
7 |
+
# python setup.py install
|
8 |
+
|
9 |
+
from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig
|
10 |
+
from peft import PeftModel, PeftConfig
|
11 |
+
import os
|
12 |
+
import torch
|
13 |
+
|
14 |
+
#import ipdb; ipdb.set_trace()
|
15 |
+
cache_dir="/workspace/asr/peft/qlora"
|
16 |
+
# TODO set this cache_dir to the path where you stored (or, want to store) llama2-70b-hf model
|
17 |
+
|
18 |
+
lazylora_dir=os.getcwd() # the path that contains 'adapter_config.json' and 'adapter_model.bin'
|
19 |
+
|
20 |
+
config = PeftConfig.from_pretrained(lazylora_dir)
|
21 |
+
|
22 |
+
tokenizer = AutoTokenizer.from_pretrained(
|
23 |
+
config.base_model_name_or_path,
|
24 |
+
cache_dir=cache_dir,
|
25 |
+
use_auth_token=True
|
26 |
+
)
|
27 |
+
|
28 |
+
bnb_config = BitsAndBytesConfig(
|
29 |
+
load_in_4bit=True,
|
30 |
+
bnb_4bit_use_double_quant=True,
|
31 |
+
bnb_4bit_quant_type='nf4',
|
32 |
+
bnb_4bit_compute_dtype=torch.bfloat16
|
33 |
+
)
|
34 |
+
|
35 |
+
model = AutoModelForCausalLM.from_pretrained(
|
36 |
+
config.base_model_name_or_path,
|
37 |
+
quantization_config=bnb_config,
|
38 |
+
device_map="auto",
|
39 |
+
cache_dir=cache_dir,
|
40 |
+
use_auth_token=True
|
41 |
+
)
|
42 |
+
#model.print_trainable_parameters()
|
43 |
+
print(sum(p.numel() for p in model.parameters()))
|
44 |
+
# 34,751,127,552 -> half-size of 70B due to 4-bit loading
|
45 |
+
|
46 |
+
model = PeftModel.from_pretrained(model, lazylora_dir)
|
47 |
+
print('after adding lazy lora parameters:')
|
48 |
+
model.print_trainable_parameters()
|
49 |
+
# trainable params: 0 || all params: 35,579,442,176 || trainable%: 0.0
|
50 |
+
|
51 |
+
|