gpu to cpu
Browse files- medrax/tools/classification.py +8 -3
- medrax/tools/generation.py +5 -2
- medrax/tools/grounding.py +5 -3
- medrax/tools/llava_med.py +14 -5
- medrax/tools/report_generation.py +4 -4
medrax/tools/classification.py
CHANGED
@@ -47,14 +47,19 @@ class ChestXRayClassifierTool(BaseTool):
|
|
47 |
)
|
48 |
args_schema: Type[BaseModel] = ChestXRayInput
|
49 |
model: xrv.models.DenseNet = None
|
50 |
-
device: Optional[
|
51 |
transform: torchvision.transforms.Compose = None
|
52 |
|
53 |
-
def __init__(self, model_name: str = "densenet121-res224-all", device: Optional[str] =
|
54 |
super().__init__()
|
|
|
|
|
|
|
55 |
self.model = xrv.models.DenseNet(weights=model_name)
|
56 |
self.model.eval()
|
57 |
-
|
|
|
|
|
58 |
self.model = self.model.to(self.device)
|
59 |
self.transform = torchvision.transforms.Compose([xrv.datasets.XRayCenterCrop()])
|
60 |
|
|
|
47 |
)
|
48 |
args_schema: Type[BaseModel] = ChestXRayInput
|
49 |
model: xrv.models.DenseNet = None
|
50 |
+
device: Optional[torch.device] = torch.device("cpu") # Default to CPU
|
51 |
transform: torchvision.transforms.Compose = None
|
52 |
|
53 |
+
def __init__(self, model_name: str = "densenet121-res224-all", device: Optional[str] = None):
|
54 |
super().__init__()
|
55 |
+
|
56 |
+
# If device is not specified, use CUDA if available, else fallback to CPU
|
57 |
+
device = device or ("cuda" if torch.cuda.is_available() else "cpu")
|
58 |
self.model = xrv.models.DenseNet(weights=model_name)
|
59 |
self.model.eval()
|
60 |
+
|
61 |
+
# Assign device based on the passed or auto-detected option
|
62 |
+
self.device = torch.device(device)
|
63 |
self.model = self.model.to(self.device)
|
64 |
self.transform = torchvision.transforms.Compose([xrv.datasets.XRayCenterCrop()])
|
65 |
|
medrax/tools/generation.py
CHANGED
@@ -61,7 +61,10 @@ class ChestXRayGeneratorTool(BaseTool):
|
|
61 |
"""Initialize the chest X-ray generator tool."""
|
62 |
super().__init__()
|
63 |
|
64 |
-
|
|
|
|
|
|
|
65 |
self.model = StableDiffusionPipeline.from_pretrained(model_path, cache_dir=cache_dir)
|
66 |
self.model = self.model.to(torch.float32).to(self.device)
|
67 |
|
@@ -121,7 +124,7 @@ class ChestXRayGeneratorTool(BaseTool):
|
|
121 |
|
122 |
except Exception as e:
|
123 |
return (
|
124 |
-
{"error": str(e)},
|
125 |
{
|
126 |
"prompt": prompt,
|
127 |
"analysis_status": "failed",
|
|
|
61 |
"""Initialize the chest X-ray generator tool."""
|
62 |
super().__init__()
|
63 |
|
64 |
+
# Automatically detect device (cuda if available, else cpu)
|
65 |
+
device = device or ("cuda" if torch.cuda.is_available() else "cpu")
|
66 |
+
self.device = torch.device(device)
|
67 |
+
|
68 |
self.model = StableDiffusionPipeline.from_pretrained(model_path, cache_dir=cache_dir)
|
69 |
self.model = self.model.to(torch.float32).to(self.device)
|
70 |
|
|
|
124 |
|
125 |
except Exception as e:
|
126 |
return (
|
127 |
+
{"error": str(e)} ,
|
128 |
{
|
129 |
"prompt": prompt,
|
130 |
"analysis_status": "failed",
|
medrax/tools/grounding.py
CHANGED
@@ -50,7 +50,7 @@ class XRayPhraseGroundingTool(BaseTool):
|
|
50 |
|
51 |
model: Any = None
|
52 |
processor: Any = None
|
53 |
-
device:
|
54 |
temp_dir: Path = None
|
55 |
|
56 |
def __init__(
|
@@ -64,7 +64,10 @@ class XRayPhraseGroundingTool(BaseTool):
|
|
64 |
):
|
65 |
"""Initialize the XRay Phrase Grounding Tool."""
|
66 |
super().__init__()
|
67 |
-
|
|
|
|
|
|
|
68 |
|
69 |
# Setup quantization config
|
70 |
if load_in_4bit:
|
@@ -93,7 +96,6 @@ class XRayPhraseGroundingTool(BaseTool):
|
|
93 |
model_path, cache_dir=cache_dir, trust_remote_code=True
|
94 |
)
|
95 |
|
96 |
-
|
97 |
self.model = self.model.eval()
|
98 |
|
99 |
self.temp_dir = Path(temp_dir if temp_dir else tempfile.mkdtemp())
|
|
|
50 |
|
51 |
model: Any = None
|
52 |
processor: Any = None
|
53 |
+
device: torch.device = None
|
54 |
temp_dir: Path = None
|
55 |
|
56 |
def __init__(
|
|
|
64 |
):
|
65 |
"""Initialize the XRay Phrase Grounding Tool."""
|
66 |
super().__init__()
|
67 |
+
|
68 |
+
# Automatically detect device (cuda if available, else cpu)
|
69 |
+
device = device or ("cuda" if torch.cuda.is_available() else "cpu")
|
70 |
+
self.device = torch.device(device)
|
71 |
|
72 |
# Setup quantization config
|
73 |
if load_in_4bit:
|
|
|
96 |
model_path, cache_dir=cache_dir, trust_remote_code=True
|
97 |
)
|
98 |
|
|
|
99 |
self.model = self.model.eval()
|
100 |
|
101 |
self.temp_dir = Path(temp_dir if temp_dir else tempfile.mkdtemp())
|
medrax/tools/llava_med.py
CHANGED
@@ -11,7 +11,6 @@ from langchain_core.tools import BaseTool
|
|
11 |
|
12 |
from PIL import Image
|
13 |
|
14 |
-
|
15 |
from medrax.llava.conversation import conv_templates
|
16 |
from medrax.llava.model.builder import load_pretrained_model
|
17 |
from medrax.llava.mm_utils import tokenizer_image_token, process_images
|
@@ -65,6 +64,11 @@ class LlavaMedTool(BaseTool):
|
|
65 |
**kwargs,
|
66 |
):
|
67 |
super().__init__()
|
|
|
|
|
|
|
|
|
|
|
68 |
self.tokenizer, self.model, self.image_processor, self.context_len = load_pretrained_model(
|
69 |
model_path=model_path,
|
70 |
model_base=None,
|
@@ -77,6 +81,9 @@ class LlavaMedTool(BaseTool):
|
|
77 |
device=device,
|
78 |
**kwargs,
|
79 |
)
|
|
|
|
|
|
|
80 |
self.model.eval()
|
81 |
|
82 |
def _process_input(
|
@@ -101,14 +108,14 @@ class LlavaMedTool(BaseTool):
|
|
101 |
input_ids = (
|
102 |
tokenizer_image_token(prompt, self.tokenizer, IMAGE_TOKEN_INDEX, return_tensors="pt")
|
103 |
.unsqueeze(0)
|
104 |
-
.
|
105 |
)
|
106 |
|
107 |
image_tensor = None
|
108 |
if image_path:
|
109 |
image = Image.open(image_path)
|
110 |
image_tensor = process_images([image], self.image_processor, self.model.config)[0]
|
111 |
-
image_tensor = image_tensor.unsqueeze(0).
|
112 |
|
113 |
return input_ids, image_tensor
|
114 |
|
@@ -133,8 +140,10 @@ class LlavaMedTool(BaseTool):
|
|
133 |
"""
|
134 |
try:
|
135 |
input_ids, image_tensor = self._process_input(question, image_path)
|
136 |
-
|
137 |
-
|
|
|
|
|
138 |
|
139 |
with torch.inference_mode():
|
140 |
output_ids = self.model.generate(
|
|
|
11 |
|
12 |
from PIL import Image
|
13 |
|
|
|
14 |
from medrax.llava.conversation import conv_templates
|
15 |
from medrax.llava.model.builder import load_pretrained_model
|
16 |
from medrax.llava.mm_utils import tokenizer_image_token, process_images
|
|
|
64 |
**kwargs,
|
65 |
):
|
66 |
super().__init__()
|
67 |
+
|
68 |
+
# Set the device (cuda or cpu)
|
69 |
+
self.device = torch.device(device) if device else torch.device("cuda")
|
70 |
+
|
71 |
+
# Load the model and tokenizer
|
72 |
self.tokenizer, self.model, self.image_processor, self.context_len = load_pretrained_model(
|
73 |
model_path=model_path,
|
74 |
model_base=None,
|
|
|
81 |
device=device,
|
82 |
**kwargs,
|
83 |
)
|
84 |
+
|
85 |
+
# Move the model to the desired device
|
86 |
+
self.model.to(self.device)
|
87 |
self.model.eval()
|
88 |
|
89 |
def _process_input(
|
|
|
108 |
input_ids = (
|
109 |
tokenizer_image_token(prompt, self.tokenizer, IMAGE_TOKEN_INDEX, return_tensors="pt")
|
110 |
.unsqueeze(0)
|
111 |
+
.to(self.device) # Move to the correct device
|
112 |
)
|
113 |
|
114 |
image_tensor = None
|
115 |
if image_path:
|
116 |
image = Image.open(image_path)
|
117 |
image_tensor = process_images([image], self.image_processor, self.model.config)[0]
|
118 |
+
image_tensor = image_tensor.unsqueeze(0).to(self.device, dtype=self.model.dtype) # Move to device
|
119 |
|
120 |
return input_ids, image_tensor
|
121 |
|
|
|
140 |
"""
|
141 |
try:
|
142 |
input_ids, image_tensor = self._process_input(question, image_path)
|
143 |
+
|
144 |
+
# Ensure that inputs are on the same device as the model
|
145 |
+
input_ids = input_ids.to(self.device)
|
146 |
+
image_tensor = image_tensor.to(self.device, dtype=self.model.dtype)
|
147 |
|
148 |
with torch.inference_mode():
|
149 |
output_ids = self.model.generate(
|
medrax/tools/report_generation.py
CHANGED
@@ -47,7 +47,7 @@ class ChestXRayReportGeneratorTool(BaseTool):
|
|
47 |
"to a chest X-ray image file. Output is a structured report with both detailed "
|
48 |
"observations and key clinical conclusions."
|
49 |
)
|
50 |
-
device: Optional[str] = "
|
51 |
args_schema: Type[BaseModel] = ChestXRayInput
|
52 |
findings_model: VisionEncoderDecoderModel = None
|
53 |
impression_model: VisionEncoderDecoderModel = None
|
@@ -57,10 +57,10 @@ class ChestXRayReportGeneratorTool(BaseTool):
|
|
57 |
impression_processor: ViTImageProcessor = None
|
58 |
generation_args: Dict[str, Any] = None
|
59 |
|
60 |
-
def __init__(self, cache_dir: str = "/model-weights", device: Optional[str] = "
|
61 |
"""Initialize the ChestXRayReportGeneratorTool with both findings and impression models."""
|
62 |
super().__init__()
|
63 |
-
self.device = torch.device(device) if device else "
|
64 |
|
65 |
# Initialize findings model
|
66 |
self.findings_model = VisionEncoderDecoderModel.from_pretrained(
|
@@ -84,7 +84,7 @@ class ChestXRayReportGeneratorTool(BaseTool):
|
|
84 |
"IAMJB/chexpert-mimic-cxr-impression-baseline", cache_dir=cache_dir
|
85 |
)
|
86 |
|
87 |
-
# Move models to device
|
88 |
self.findings_model = self.findings_model.to(self.device)
|
89 |
self.impression_model = self.impression_model.to(self.device)
|
90 |
|
|
|
47 |
"to a chest X-ray image file. Output is a structured report with both detailed "
|
48 |
"observations and key clinical conclusions."
|
49 |
)
|
50 |
+
device: Optional[str] = "cpu" # Change the device to "cpu"
|
51 |
args_schema: Type[BaseModel] = ChestXRayInput
|
52 |
findings_model: VisionEncoderDecoderModel = None
|
53 |
impression_model: VisionEncoderDecoderModel = None
|
|
|
57 |
impression_processor: ViTImageProcessor = None
|
58 |
generation_args: Dict[str, Any] = None
|
59 |
|
60 |
+
def __init__(self, cache_dir: str = "/model-weights", device: Optional[str] = "cpu"):
|
61 |
"""Initialize the ChestXRayReportGeneratorTool with both findings and impression models."""
|
62 |
super().__init__()
|
63 |
+
self.device = torch.device(device) if device else torch.device("cpu") # Ensure CPU is used
|
64 |
|
65 |
# Initialize findings model
|
66 |
self.findings_model = VisionEncoderDecoderModel.from_pretrained(
|
|
|
84 |
"IAMJB/chexpert-mimic-cxr-impression-baseline", cache_dir=cache_dir
|
85 |
)
|
86 |
|
87 |
+
# Move models to device (CPU)
|
88 |
self.findings_model = self.findings_model.to(self.device)
|
89 |
self.impression_model = self.impression_model.to(self.device)
|
90 |
|