[add] display result in tables.

.gitignore

@@ -160,3 +160,6 @@ cython_debug/
 #  and can be added to the global gitignore or merged into this file.  For a more nuclear
 #  option (not recommended) you can uncomment the following to ignore the entire idea folder.
 #.idea/
+
+# Custom
+gguf-parser-*

.vscode/settings.json

@@ -0,0 +1,11 @@
+{
+    "[python]": {
+        "editor.defaultFormatter": "ms-python.black-formatter",
+        "editor.formatOnSave": true,
+        "editor.codeActionsOnSave": {
+            "source.organizeImports": true
+        },
+      },
+    "isort.args":["--profile", "black"],
+    
+}

app.py

@@ -1,21 +1,78 @@
+import json
 import os
 from pathlib import Path
 
 import gradio as gr
+import pandas as pd
 
 GGUF_PARSER_VERSION = os.getenv("GGUF_PARSER_VERSION", "v0.12.0")
 gguf_parser = Path("gguf-parser-linux-amd64")
+gguf_parser_url = f"https://github.com/gpustack/gguf-parser-go/releases/download/{GGUF_PARSER_VERSION}/{gguf_parser}"
 
-def greet(tmp):
-    # Run the gguf-parser-go binary
-    gguf_parser_output = os.popen(f"./{gguf_parser} --version").read()
-    return f"{gguf_parser_output}"
 
-iface = gr.Interface(fn=greet, inputs="text", outputs="text")
+def process_url(url):
+    try:
+        res = os.popen(f"./{gguf_parser} -url {url} --json").read()
+        data = json.loads(res)
+
+        architecture_df = pd.DataFrame([data["architecture"]])
+
+        estimate_df = pd.DataFrame(
+            [
+                {
+                    # "maximumTokensPerSecond": data["estimate"]["items"][0][
+                    #     "maximumTokensPerSecond"
+                    # ],
+                    "offloadLayers": data["estimate"]["items"][0]["offloadLayers"],
+                    "fullOffloaded": data["estimate"]["items"][0]["fullOffloaded"],
+                    "contextSize": data["estimate"]["contextSize"],
+                    "flashAttention": data["estimate"]["flashAttention"],
+                    "distributable": data["estimate"]["distributable"],
+                }
+            ]
+        )
+
+        metadata_df = pd.DataFrame([data["metadata"]])
+
+        tokenizer_df = pd.DataFrame([data["tokenizer"]])
+
+        return architecture_df, estimate_df, metadata_df, tokenizer_df
+    except Exception as e:
+        return e
+
 
 if __name__ == "__main__":
     if not gguf_parser.exists():
-        gguf_parser_url = f"https://github.com/gpustack/gguf-parser-go/releases/download/{GGUF_PARSER_VERSION}/gguf-parser-linux-amd64"
         os.system(f"wget {gguf_parser_url}")
         os.system(f"chmod +x {gguf_parser}")
+
+    with open("devices.json", "r", encoding="utf-8") as f:
+        device_list = json.load(f)
+
+    with gr.Blocks(title="GGUF 分析器") as iface:
+        url_input = gr.Textbox(label="輸入 GGUF URL")
+        submit_btn = gr.Button("送出")
+
+        gr.Markdown("### 模型架構")
+        architecture_table = gr.DataFrame()
+
+        gr.Markdown("### 效能評估")
+        estimate_table = gr.DataFrame()
+
+        gr.Markdown("### 中繼資料")
+        metadata_table = gr.DataFrame()
+
+        gr.Markdown("### 分詞器")
+        tokenizer_table = gr.DataFrame()
+
+        submit_btn.click(
+            fn=process_url,
+            inputs=url_input,
+            outputs=[
+                architecture_table,
+                estimate_table,
+                metadata_table,
+                tokenizer_table,
+            ],
+        )
     iface.launch()

devices.json

@@ -0,0 +1,15 @@
+[
+    {"model": "V100", "memory-size": 32, "memory-bandwidth": 900, "FLOPS": "112.224TFLOPS"},
+    {"model": "T4", "memory-size": 16, "memory-bandwidth": 320, "FLOPS": "64.8TFLOPS"},
+    {"model": "A2", "memory-size": 16, "memory-bandwidth": 200, "FLOPS": "18.124TFLOPS"},
+    {"model": "A10", "memory-size": 24, "memory-bandwidth": 600, "FLOPS": "124.96TFLOPS"},
+    {"model": "A16*4", "memory-size": 64, "memory-bandwidth": 800, "FLOPS": "73.728TFLOPS"},
+    {"model": "A30", "memory-size": 24, "memory-bandwidth": 933.1, "FLOPS": "165.12TFLOPS"},
+    {"model": "A40", "memory-size": 48, "memory-bandwidth": 695.8, "FLOPS": "149.68TFLOPS"},
+    {"model": "A100-40GB", "memory-size": 40, "memory-bandwidth": 1555, "FLOPS": "312.0TFLOPS"},
+    {"model": "A100-80GB", "memory-size": 80, "memory-bandwidth": 1555, "FLOPS": "312.0TFLOPS"},
+    {"model": "H100-PCIE", "memory-size": 80, "memory-bandwidth": 2039, "FLOPS": "756.449TFLOPS"},
+    {"model": "H100-SXM", "memory-size": 80, "memory-bandwidth": 3352, "FLOPS": "989.43TFLOPS"},
+    {"model": "L40", "memory-size": 48, "memory-bandwidth": 864, "FLOPS": "362.066TFLOPS"},
+    {"model": "L4", "memory-size": 24, "memory-bandwidth": 300, "FLOPS": "121.0TFLOPS"}
+]

pyproject.toml

@@ -7,4 +7,8 @@ requires-python = ">=3.10"
 dependencies = [
     "gradio==5.3.0",
     "orjson==3.10.10",
+    "pandas>=2.2.3",
 ]
+
+[tool.pylint]
+max-line-length = 180

uv.lock

@@ -196,12 +196,14 @@ source = { virtual = "." }
 dependencies = [
     { name = "gradio" },
     { name = "orjson" },
+    { name = "pandas" },
 ]
 
 [package.metadata]
 requires-dist = [
     { name = "gradio", specifier = "==5.3.0" },
     { name = "orjson", specifier = "==3.10.10" },
+    { name = "pandas", specifier = ">=2.2.3" },
 ]
 
 [[package]]