Update README.md
Browse files
README.md
CHANGED
|
@@ -6,7 +6,7 @@ parent_dir/
|
|
| 6 |
├── Q-RAG/ ← [Q-RAG](https://github.com/griver/Q-RAG.git)
|
| 7 |
└── datasets/ ← [datasets Hotpotqa and Musique](https://huggingface.co/datasets/Q-RAG/Hotpotqa_and_Musique)
|
| 8 |
```
|
| 9 |
-
Git datasets for Q-RAG
|
| 10 |
```bash
|
| 11 |
git clone https://huggingface.co/datasets/Q-RAG/Hotpotqa_and_Musique
|
| 12 |
cd Hotpotqa_and_Musique
|
|
@@ -15,14 +15,14 @@ cd ..
|
|
| 15 |
rm -rf Hotpotqa_and_Musique
|
| 16 |
du -h
|
| 17 |
```
|
| 18 |
-
Git repo of Q-RAG
|
| 19 |
```bash
|
| 20 |
git clone https://github.com/griver/Q-RAG.git
|
| 21 |
cd Q-RAG
|
| 22 |
#Only need when you don't have your self-trained hotpotqa model yet
|
| 23 |
git clone https://huggingface.co/Q-RAG/qrag-ft-e5-on-hotpotqa
|
| 24 |
```
|
| 25 |
-
Environment Setup
|
| 26 |
```bash
|
| 27 |
# Setup venv
|
| 28 |
conda create -n qrag python=3.12 -y
|
|
@@ -36,7 +36,7 @@ pip install hydra-core tensorboard rotary-embedding-torch pandas nltk sortedcont
|
|
| 36 |
python -c "from rl.agents.pqn import PQNActor; print('✅ Q-RAG installed successfully')"
|
| 37 |
|
| 38 |
```
|
| 39 |
-
Train: Log with Time
|
| 40 |
```bash
|
| 41 |
python train_q_rag_logt.py \
|
| 42 |
envs=hotpotqa \
|
|
@@ -49,7 +49,7 @@ python train_q_rag_logt.py \
|
|
| 49 |
envs_parallel=1 \
|
| 50 |
max_action_length=220
|
| 51 |
```
|
| 52 |
-
Original Train
|
| 53 |
```bash
|
| 54 |
python train_q_rag.py \
|
| 55 |
envs=hotpotqa \
|
|
@@ -62,4 +62,17 @@ python train_q_rag.py \
|
|
| 62 |
envs_parallel=1 \
|
| 63 |
max_action_length=220
|
| 64 |
|
| 65 |
-
```
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 6 |
├── Q-RAG/ ← [Q-RAG](https://github.com/griver/Q-RAG.git)
|
| 7 |
└── datasets/ ← [datasets Hotpotqa and Musique](https://huggingface.co/datasets/Q-RAG/Hotpotqa_and_Musique)
|
| 8 |
```
|
| 9 |
+
### Git datasets for Q-RAG
|
| 10 |
```bash
|
| 11 |
git clone https://huggingface.co/datasets/Q-RAG/Hotpotqa_and_Musique
|
| 12 |
cd Hotpotqa_and_Musique
|
|
|
|
| 15 |
rm -rf Hotpotqa_and_Musique
|
| 16 |
du -h
|
| 17 |
```
|
| 18 |
+
### Git repo of Q-RAG
|
| 19 |
```bash
|
| 20 |
git clone https://github.com/griver/Q-RAG.git
|
| 21 |
cd Q-RAG
|
| 22 |
#Only need when you don't have your self-trained hotpotqa model yet
|
| 23 |
git clone https://huggingface.co/Q-RAG/qrag-ft-e5-on-hotpotqa
|
| 24 |
```
|
| 25 |
+
### Environment Setup
|
| 26 |
```bash
|
| 27 |
# Setup venv
|
| 28 |
conda create -n qrag python=3.12 -y
|
|
|
|
| 36 |
python -c "from rl.agents.pqn import PQNActor; print('✅ Q-RAG installed successfully')"
|
| 37 |
|
| 38 |
```
|
| 39 |
+
### Train: Log with Time
|
| 40 |
```bash
|
| 41 |
python train_q_rag_logt.py \
|
| 42 |
envs=hotpotqa \
|
|
|
|
| 49 |
envs_parallel=1 \
|
| 50 |
max_action_length=220
|
| 51 |
```
|
| 52 |
+
### Original Train
|
| 53 |
```bash
|
| 54 |
python train_q_rag.py \
|
| 55 |
envs=hotpotqa \
|
|
|
|
| 62 |
envs_parallel=1 \
|
| 63 |
max_action_length=220
|
| 64 |
|
| 65 |
+
```
|
| 66 |
+
|
| 67 |
+
## Computer resources
|
| 68 |
+
[基于HotpotQA+Musique(combined, GTE embedder) 训练出来的模型](QRAG_combined.zip) Q-RAG文中没有提及他的测试 <br>
|
| 69 |
+
- 训练时常：18:07:48
|
| 70 |
+
- 显卡： Pro 6000 96GB
|
| 71 |
+
- 显存占用：60GB ± 0.5GB
|
| 72 |
+
|
| 73 |
+
|
| 74 |
+
HotpotQA_推理
|
| 75 |
+
- 训练时常：00:12:26
|
| 76 |
+
- 显卡：NVIDIA A100-SXM4-80GB
|
| 77 |
+
- 显存占用：30GB ± 1GB
|
| 78 |
+
|