GirishaBuilds01 commited on
Commit
4993b60
·
verified ·
1 Parent(s): 004763d

Delete app.py

Browse files
Files changed (1) hide show
  1. app.py +0 -55
app.py DELETED
@@ -1,55 +0,0 @@
1
- import gradio as gr
2
- import pandas as pd
3
- import torch
4
-
5
- from core.model_loader import load_model, SUPPORTED_MODELS
6
- from core.profiler import ActivationProfiler
7
- from core.stress import stress_layer
8
- from core.sensitivity import compute_si, assign_policy
9
- from core.reevaluator import reevaluate
10
- from core.visualization import create_heatmap
11
-
12
- def run_cortex(model_choice, calibration_text):
13
- try:
14
- model, tokenizer = load_model(model_choice)
15
-
16
- # Fix GPT2 padding issue
17
- if tokenizer.pad_token is None:
18
- tokenizer.pad_token = tokenizer.eos_token
19
-
20
- texts = calibration_text.strip().split("\n")
21
- texts = [t.strip() for t in texts if t.strip() != ""]
22
-
23
- if len(texts) == 0:
24
- return {"error": "No valid input sentences provided."}, None, 0.0, None
25
-
26
- inputs = tokenizer(texts, return_tensors="pt", padding=True, truncation=True)
27
-
28
- profiler = ActivationProfiler(model)
29
- profiler.register()
30
-
31
- with torch.no_grad():
32
- model(**inputs)
33
-
34
- profiler.remove()
35
- stats = profiler.get()
36
-
37
- stress_results = {}
38
-
39
- # 🔥 LIMIT layers for stability (important for HF CPU)
40
- for name in list(stats.keys())[:6]:
41
- mse_val, kl_val = stress_layer(model, name, inputs, bits=8)
42
- stress_results[name] = (mse_val, kl_val)
43
-
44
- scores = compute_si(stats, stress_results)
45
- policy = assign_policy(scores)
46
-
47
- policy, degradation = reevaluate(model, tokenizer, texts, policy)
48
-
49
- heatmap = create_heatmap(scores)
50
-
51
- return policy, heatmap, float(degradation), "layer_metrics.csv"
52
-
53
- except Exception as e:
54
- return {"error": str(e)}, None, 0.0, None
55
-