Fix autoawq kernels requirements

README.md

@@ -11,11 +11,12 @@ Create a new conda environment for idefics2. This requirements file suppose that
 
 ```bash
 conda create -n idefics2 python=3.10
+conda activate idefics2
 pip install -r requirements.txt
 pip install -e <PATH TO LATEST TRANSFOMERS VERSION>
 ```
 
-## Running the demo
+## Robomaster Connection
 
 Make sure to be connected using the wifi hotspot of the robomaster which is the most stable one.
 
@@ -23,6 +24,24 @@ The default password for the hotpsot is: 12341234
 
 You might need to have a second wifi card if you want to be able to run the demo with internet on.
 
+## Post-Installation test
+
+Please try running idefics2 with:
+
+```bash
+conda activate idefics2
+python tests/test_idefics2.py
+```
+
+Please try running robomaster with:
+
+```bash
+conda activate robomaster
+python tests/test_robomaster.py
+```
+
+## Running the demo
+
 ```bash
 export HF_TOKEN=<TOKEN>
 conda activate idefics2

requirements.txt

@@ -0,0 +1,11 @@
+dora-rs
+torch==2.2.0
+flash_attn @ https://github.com/Dao-AILab/flash-attention/releases/download/v2.5.6/flash_attn-2.5.6+cu122torch2.2cxx11abiFALSE-cp310-cp310-linux_x86_64.whl
+autoawq 
+autoawq-kernels
+sounddevice
+openai-whisper
+pynput
+opencv-python
+Pillow
+transformers

tests/test_idefics2.py

@@ -4,7 +4,7 @@ from PIL import Image
 from io import BytesIO
 
 from transformers import AutoProcessor, AutoModelForVision2Seq, AwqConfig
-
+import awq_ext
 
 MODE = "quantized"
 DEVICE = "cuda"
@@ -98,7 +98,6 @@ def ask_vlm(image, instruction):
 
 import time
 
-model.eval()
 now = time.time()
 print(ask_vlm(image1, "What is this?")[0].split("\nAssistant: ")[1])