aakothari commited on
Commit
33d19fb
·
verified ·
1 Parent(s): 251dd8f

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +28 -18
README.md CHANGED
@@ -3,10 +3,8 @@ title: Conformer
3
  emoji: 🧬
4
  colorFrom: gray
5
  colorTo: blue
6
- sdk: gradio
7
- sdk_version: 5.49.1
8
- python_version: '3.12'
9
- app_file: app.py
10
  pinned: false
11
  license: mit
12
  short_description: BRAID molecular workbench with BRAIDBERTa
@@ -20,23 +18,35 @@ structures, compute descriptors, benchmark tokenisations, embed and compare
20
  encoders, export datasets.
21
 
22
  Nothing to install. Structure parsing, drawing, descriptors and fingerprints run
23
- in your browser via RDKit's WebAssembly build. BRAID conversion and the encoders
24
- run on this Space, on CPU.
 
 
 
 
 
 
 
 
 
 
 
25
 
26
  - Code: https://github.com/AayushK-othari/braid
27
- - Model: https://huggingface.co/aakothari/BRAIDBERTa
 
 
 
 
28
 
29
- `GET /version` returns the git SHA, pinned model revisions and package versions
30
- behind whatever you are looking at.
31
 
32
- ## A note on `sdk_version`
 
33
 
34
- The Space builder appends `gradio==<sdk_version>` and `spaces==<version>` to the
35
- `pip install` line it runs against `requirements.txt`. Those two packages are
36
- therefore deliberately absent from `requirements.txt` — pinning them in both
37
- places yields two conflicting `==` constraints and the build fails to resolve.
38
 
39
- `5.49.1` is not arbitrary. Gradio 6.x requires `huggingface-hub>=1.2`, which
40
- `transformers==4.49.0` forbids (`<1.0`). Gradio 5.x is the branch that coexists
41
- with the transformers 4 stack these results were produced on. Changing this
42
- number means changing the encoder stack, so re-validate locally before you do.
 
3
  emoji: 🧬
4
  colorFrom: gray
5
  colorTo: blue
6
+ sdk: static
7
+ app_file: index.html
 
 
8
  pinned: false
9
  license: mit
10
  short_description: BRAID molecular workbench with BRAIDBERTa
 
18
  encoders, export datasets.
19
 
20
  Nothing to install. Structure parsing, drawing, descriptors and fingerprints run
21
+ entirely in your browser via RDKit's WebAssembly build, so most of the workbench
22
+ works with no server at all.
23
+
24
+ BRAID conversion and the encoders need Python — `braids.codec` depends on RDKit
25
+ APIs that the WebAssembly build does not expose — so those routes are served by
26
+ a small FastAPI service:
27
+
28
+ **https://braid-9wc2.onrender.com**
29
+
30
+ The encoders run there from ONNX graphs exported from the checkpoints below,
31
+ on CPU. That service is on a free instance and sleeps when idle: the first
32
+ request after a quiet spell can take ~50 seconds to wake it. Everything
33
+ browser-side stays instant meanwhile.
34
 
35
  - Code: https://github.com/AayushK-othari/braid
36
+ - Models: https://huggingface.co/aakothari/BRAIDBERTa
37
+ and https://huggingface.co/aakothari/DeepBERTa_zinc_base_100k_v4
38
+
39
+ `GET /version` on the backend returns the git SHA, the pinned model revisions
40
+ and the installed package versions behind whatever you are looking at.
41
 
42
+ ## Pointing at your own backend
 
43
 
44
+ The workbench's settings panel takes any endpoint. To run the converters and
45
+ encoders locally instead — nothing leaves your machine:
46
 
47
+ ```bash
48
+ pip install -r requirements-backend.txt
49
+ python backend.py
50
+ ```
51
 
52
+ then set the endpoint to `http://127.0.0.1:8000`.