bithal26 commited on
Commit
44894ea
Β·
verified Β·
1 Parent(s): 845ba84

Update requirements.txt

Browse files
Files changed (1) hide show
  1. requirements.txt +14 -26
requirements.txt CHANGED
@@ -1,32 +1,20 @@
1
  # ──────────────────────────────────────────────────────────────────────────────
2
  # VERIDEX β€” Worker Space requirements.txt
3
  #
4
- # PINNING RULES FOR HF SPACES
5
- # ────────────────────────────
6
- # 1. NEVER pin gradio β€” HF force-installs gradio==6.x in [base 6/7].
7
- # Any pin causes a ResolutionImpossible error β†’ exit code 1.
8
- #
9
- # 2. NEVER hard-pin torch/torchvision to a specific patch version.
10
- # The HF base image ships a pre-built torch; an exact pin that doesn't
11
- # match the available wheel causes "No matching distribution found".
12
- # Use a floor (>=) so pip can pick the pre-installed wheel.
13
- #
14
- # 3. NEVER pin timm to 0.9.x β€” timm 1.x removed the direct
15
- # timm.models.efficientnet.tf_efficientnet_b7_ns import path.
16
- # Our app.py already has a try/except fallback, but a hard pin
17
- # to 0.9.x will conflict with timm 1.x that Gradio 6 may pull in.
18
- # Use a floor instead; the fallback import handles both versions.
19
- #
20
- # 4. numpy: Gradio 6 requires numpy>=1.0,<3.0. numpy 1.26.x is fine but
21
- # do not hard-pin it β€” let pip choose within that window.
22
  # ──────────────────────────────────────────────────────────────────────────────
23
 
24
- # Deep-learning stack (floor-only β€” let HF pick the compatible wheel)
25
- torch>=2.1.0
26
- torchvision>=0.16.0
27
-
28
- # Model architecture
29
- timm>=0.9.16
 
30
 
31
- # Numerical
32
- numpy>=1.24,<3.0
 
1
  # ──────────────────────────────────────────────────────────────────────────────
2
  # VERIDEX β€” Worker Space requirements.txt
3
  #
4
+ # HF Spaces runs Python 3.13. Rules:
5
+ # 1. Never pin gradio β€” HF force-installs gradio==6.x in [base 6/7].
6
+ # 2. Use floor-only constraints so pip picks available cp313 wheels.
7
+ # 3. numpy β‰₯ 2.0 β€” first release with a cp313 wheel (1.26.x source-builds
8
+ # and fails on Python 3.13).
 
 
 
 
 
 
 
 
 
 
 
 
 
9
  # ──────────────────────────────────────────────────────────────────────────────
10
 
11
+ # ── Python 3.13 WHEEL-SAFE floors (verified on PyPI cp313-cp313-manylinux) ───
12
+ # torch 2.6.0 = first release with a cp313-cp313-manylinux_2_28 wheel
13
+ # torchvision 0.21.0 = paired release for torch 2.6
14
+ # numpy 2.0 = first cp313 wheel
15
+ torch>=2.6.0
16
+ torchvision>=0.21.0
17
+ numpy>=2.0
18
 
19
+ # Model architecture (pure-Python wheel, any version installs fine on Py3.13)
20
+ timm>=1.0.0