Kashaf1 commited on
Commit
d392000
·
1 Parent(s): 84e62d7

Uploaded complete moirai_forecast_app contents

Browse files
Files changed (1) hide show
  1. requirements.txt +19 -13
requirements.txt CHANGED
@@ -6,8 +6,7 @@ gluonts>=0.14
6
  # Data
7
  yfinance>=0.2.40
8
  pandas>=2.0
9
- numpy>=1.24,<2.0
10
- scipy>=1.11
11
 
12
  # UI (Hugging Face Spaces)
13
  gradio>=4.40
@@ -15,14 +14,21 @@ gradio>=4.40
15
  python-dateutil
16
  pytz
17
 
18
- # --- KNOWN COMPATIBILITY RISK (read before deploying) ---
19
- # This build was written and unit-tested in a sandbox running numpy 2.4.4
20
- # and pandas 3.0.2 — newer than the numpy<2 pin uni2ts's own dependency
21
- # chain (jax/lightning) has historically required. `pip install uni2ts`
22
- # will resolve numpy down to whatever it needs; the numpy<2 pin above
23
- # reflects that expectation. If the uni2ts release you install supports
24
- # numpy 2.x, relax this pin. Verify with:
25
- # pip install -r requirements.txt
26
- # python -c "from uni2ts.model.moirai_moe import MoiraiMoEModule; \
27
- # MoiraiMoEModule.from_pretrained('Salesforce/moirai-moe-1.0-R-small')"
28
- # before trusting any forecast this app produces (spec section 75).
 
 
 
 
 
 
 
 
6
  # Data
7
  yfinance>=0.2.40
8
  pandas>=2.0
9
+ numpy>=1.24
 
10
 
11
  # UI (Hugging Face Spaces)
12
  gradio>=4.40
 
14
  python-dateutil
15
  pytz
16
 
17
+ # --- FIX applied after a real HF Spaces build failure ---
18
+ # An earlier version of this file pinned `numpy<2.0` and an explicit
19
+ # `scipy>=1.11` as an untested precaution. On a real Spaces build this
20
+ # forced pip to resolve an old scipy release with no prebuilt wheel for
21
+ # the Space's Python version, which then failed trying to build scipy
22
+ # from source via meson (no Fortran compiler / BLAS-LAPACK toolchain in
23
+ # that install step -> "metadata-generation-failed"). Neither pin was
24
+ # load-bearing: nothing in this codebase imports scipy directly, and
25
+ # none of our own numpy usage requires numpy<2. Removing both lets
26
+ # uni2ts/gluonts/torch resolve numpy (and scipy, transitively) to
27
+ # whatever versions they actually ship wheels for.
28
+ #
29
+ # General pattern if a DIFFERENT package fails "metadata-generation" or
30
+ # tries to build from source on Spaces: it's almost always pip being
31
+ # forced into a source build for a package with no wheel for that
32
+ # Python version. Fix is usually the same shape -- stop pinning that
33
+ # package explicitly (or bump its floor) and let its real dependent
34
+ # choose a version it has a wheel for.