Skip to content

Instantly share code, notes, and snippets.

View fangkuoyu's full-sized avatar

fangkuoyu

1 follower · 0 following
View GitHub Profile
@fangkuoyu
fangkuoyu / eval.csv
Last active December 15, 2024 10:10
evalulation
google/paligemma-3B-pt-224 google/paligemma-3B-mix-224 google/paligemma2-3B-pt-224
BERTScore (acc/avg) G-Eval (acc/avg) BERTScore (acc/avg) G-Eval (acc/avg) BERTScore (acc/avg) G-Eval (acc/avg)
VQA-RAD Train Data 200
VQA-RAD Test Data 200
@fangkuoyu
fangkuoyu / PaliGemma.csv
Last active December 10, 2024 02:23
PaliGemma
PaliGemma v1 PaliGemma v2
Model Size 3B 3B / 10B / 28B
Training Mode Pre-Trained(pt) / Fine-Tuned (mix) Pre-Trained(pt)
Image Resolution 224x224 / 448x448 / 896x896 224x224 / 448x448 / 896x896
Release Example paligemma-3b-pt-224 paligemma2-3b-pt-224
paligemma-3b-mix-224 paligemma2-28b-pt-896
@fangkuoyu
fangkuoyu / TinyLLaMA Finetuning & Alignment.csv
Last active April 7, 2024 15:39
Model Data Training Method Library Colab Local
Transformer PEFT TRL
TinyLLaMA Finetuning PEFT STFTrainer
LoRA SFTTrainer
QLoRA SFTTrainer T4 1660Ti
Alignment RLHF/PPO PPOTrainer
DPO DPOTrainer 1660Ti
GPT2 IMDB Aligment RLHF/PPO yes no PPOTrainer T4
@fangkuoyu
fangkuoyu / RL_env_2.csv
Last active March 29, 2023 13:39
We can make this file beautiful and searchable if this error is corrected: It looks like row 10 should actually have 10 columns, instead of 2 in line 9.
Environment, OpenAI Gym, D/C, Self-Driving, Q-Learning, DQN, DDPG, Policy Gradient, A2C/A3C, PPO
PyLessons - Pong, , , , , pong_dqn_tf2_001.py/222/-4.48, , pong_pg_tf2_001/300/-18 TBD, v, TBD
PyLessons - LunarLander, , , , , , , , , v
Jordi Torres, , , , , , , , ,
, , , , , , ,,,
CartPole, v, , , , , ,,,
Pendulum, v, C, , , , ,,,
LunarLander, v, D, v, , NG/TBD, ,,,
LunarLander, v, C, v, , , SKRL/500000/failed, ,,?
CarRacing,
@fangkuoyu
fangkuoyu / RL_env.csv
Last active December 24, 2022 03:52
Name Pole Taxi Atari/Pong Atari/Breakout
State Space of Variables 4 1 (210, 160, 3) (210, 160, 3)
Range of Variable float int (1-500) int int
Action 2 6 6 6
Reward
Goal up pick-drop return hit the brick
Model Q Learning v
Deep Q Learning v v v v
Comment 5 lines
@fangkuoyu
fangkuoyu / wms_dataset.csv
Last active November 23, 2022 09:01
Type Star Title Description Paper_URL Dataset_URL
(arxiv) (Kaggle/Zenodo)
Networking
WiFi Toward More Reliable Deep Learning-Based Link Adaptation for WiFi 6
SMART - Self-adaptive Machine Learning Approach for Real-time Tuning of IEEE 802.11 PHY and MAC layers https://zenodo.org/record/6394799#.Y3wtZ3ZBxPY
LoRa LoED: The LoRaWAN at the Edge Dataset https://zenodo.org/record/4121430#.Y3wzU3ZBxPY
* Environmental Impact on the Long-Term Connectivity and Link Quality of an Outdoor LoRa Network
NB-IoT Deep Learning Anomaly Detection for Cellular IoT With Applications in Smart Logistics https://arxiv.org/abs/2102.08936 https://zenodo.org/record/4686782#.Y3w2OHZBxPY
3G/4G/5G
* Anomaly detection in 4G cellular networks https://www.kaggle.com/competitions/anomaly-detection-in-4g-cellular-networks
@fangkuoyu
fangkuoyu / yolov7_program.csv
Last active October 17, 2022 13:06
File Description arguments outputs
1. train.py train the data defined in 'train' under pcb.yaml
2. test.py test the data defined in 'test' under pcb.yaml
3. detect.py detect an image
@fangkuoyu
fangkuoyu / pcb_test.csv
Created October 17, 2022 07:12
Class Labels P R [email protected] [email protected]:.95
all 470 0.965 0.859 0.937 0.461
Missing_hole 77 0.974 0.974 0.958 0.513
Mouse_bite 75 0.984 0.833 0.971 0.494
Open_circuit 80 0.953 0.754 0.899 0.393
Short 81 0.987 0.958 0.972 0.426
Spur 76 1 0.82 0.927 0.473
Spurious_copper 81 0.892 0.815 0.898 0.468
@fangkuoyu
fangkuoyu / pcb.yaml
Last active October 21, 2022 03:26
# train/valid/test directories
train: D:/pcb/dataset/train/
val: D:/pcb/dataset/valid/
test: D:/pcb/dataset/test/
# number of classes
nc: 6
# names of classes
names: ['Missing_hole', 'Mouse_bite', 'Open_circuit', 'Short', 'Spur', 'Spurious_copper']
@fangkuoyu
fangkuoyu / ResNet.csv
Last active August 2, 2022 04:50
ID Dataset Type Classes Total Per Class ImageNet Trainable Training Model Reg Aug Epoch Train Valid Comment
*1 Archietcture more 25 200+ - 600+ yes true true - - - 30 100 70+ fair
2 Archietcture2 more 9 900+ - 1700+ yes true true - - - 30 100 70+ fair
*3 _art/artist more 50 300+ yes true true - - - 30 100 70+ fair
*4 _art/artform more 5 700+ - 2000+ yes true true - - - 30 100 96.1 great
*5 _art/artmovement more 3 320 yes true true - - - 30 100 74.0 fair
50 100 87.5 good
*6 _house/furniture one 5 900+ - 1600+ yes true true - - - 30 100 100 gr