Pin sdk_version to 5.49.1; let the builder own gradio

README.md

@@ -1,15 +1,42 @@
 ---
 title: Conformer
-emoji: 📈
+emoji: 🧬
 colorFrom: gray
-colorTo: gray
+colorTo: blue
 sdk: gradio
-sdk_version: 6.20.0
+sdk_version: 5.49.1
 python_version: '3.12'
 app_file: app.py
 pinned: false
 license: mit
-short_description: 'AI Molecular workbench '
+short_description: BRAID molecular workbench with BRAIDBERTa
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+# Conformer
+
+A browser workbench for the BRAID molecular line notation and the BRAIDBERTa /
+DeepBERTa encoders. Load molecules, convert between representations, draw
+structures, compute descriptors, benchmark tokenisations, embed and compare
+encoders, export datasets.
+
+Nothing to install. Structure parsing, drawing, descriptors and fingerprints run
+in your browser via RDKit's WebAssembly build. BRAID conversion and the encoders
+run on this Space, on CPU.
+
+- Code: https://github.com/AayushK-othari/braid
+- Model: https://huggingface.co/aakothari/BRAIDBERTa
+
+`GET /version` returns the git SHA, pinned model revisions and package versions
+behind whatever you are looking at.
+
+## A note on `sdk_version`
+
+The Space builder appends `gradio==<sdk_version>` and `spaces==<version>` to the
+`pip install` line it runs against `requirements.txt`. Those two packages are
+therefore deliberately absent from `requirements.txt` — pinning them in both
+places yields two conflicting `==` constraints and the build fails to resolve.
+
+`5.49.1` is not arbitrary. Gradio 6.x requires `huggingface-hub>=1.2`, which
+`transformers==4.49.0` forbids (`<1.0`). Gradio 5.x is the branch that coexists
+with the transformers 4 stack these results were produced on. Changing this
+number means changing the encoder stack, so re-validate locally before you do.

requirements.txt

@@ -1,18 +1,25 @@
 # Pinned to the stack this project was actually validated against.
 #
-# Two constraints shape these versions, both discovered the hard way:
+# Three constraints shape these versions, all discovered the hard way:
 #
 #   1. ZeroGPU accepts only torch 2.11.0, 2.10.0, 2.9.1 or 2.8.0.
 #   2. gradio 6.x requires huggingface-hub >= 1.2, which transformers 4.x
 #      forbids (< 1.0). gradio 5.x is the branch that coexists with the
 #      transformers 4.49 stack running locally.
+#   3. gradio and spaces are NOT pinned here. The Space builder appends
+#      `gradio==<sdk_version from README.md>` and `spaces==<version>` to its
+#      own pip install line. Pinning either one here as well gives pip two
+#      conflicting `==` constraints on one package and the build dies with
+#      ResolutionImpossible. The gradio version is controlled by
+#      `sdk_version` in README.md and nowhere else.
 #
 # Do NOT bump transformers to 5.x here without re-running the validation
 # locally first. Deploying a different major version than the one that
 # produced your numbers is how a discrepancy becomes unexplainable.
-
-# --- Space wrapper (app.py only; backend.py never imports it) --------------
-gradio==5.49.1
+#
+# Running outside a Space (Cloud Run, Docker, local): add `gradio==5.49.1`
+# back if you are launching app.py, or skip it entirely and run backend.py,
+# which never imports gradio.
 
 # --- web layer -------------------------------------------------------------
 fastapi==0.139.2
@@ -26,8 +33,9 @@ braids-mol @ git+https://github.com/AayushK-othari/braid.git@main
 
 # --- encoders --------------------------------------------------------------
 # No CPU-only index here: ZeroGPU expects the standard CUDA build. The app is
-# CPU-bound and simply never requests a GPU.
+# CPU-bound and simply never requests a GPU. The Space base image already
+# ships torch 2.8.0, so this pin matches rather than triggers a reinstall.
 torch==2.8.0
 transformers==4.49.0
 tokenizers==0.21.0
-huggingface-hub==0.36.2
+huggingface-hub==0.36.2