Document ZeroGPU deploy and gated-model troubleshooting
Browse files
README.md
CHANGED
|
@@ -48,19 +48,28 @@ If the Space already exists on `cpu-basic`, switch hardware:
|
|
| 48 |
hf spaces settings build-small-hackathon/tiny-aya-compare --hardware zero-a10g
|
| 49 |
```
|
| 50 |
|
| 51 |
-
### 2.
|
| 52 |
|
| 53 |
-
Both
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 54 |
|
| 55 |
```bash
|
| 56 |
-
hf spaces secrets add build-small-hackathon/tiny-aya-compare
|
| 57 |
-
--key HF_TOKEN \
|
| 58 |
-
--value "$(hf auth token)"
|
| 59 |
```
|
| 60 |
|
| 61 |
-
|
|
|
|
|
|
|
| 62 |
|
| 63 |
-
###
|
| 64 |
|
| 65 |
From this repo:
|
| 66 |
|
|
@@ -81,7 +90,7 @@ hf upload build-small-hackathon/tiny-aya-compare . \
|
|
| 81 |
--include "requirements.txt"
|
| 82 |
```
|
| 83 |
|
| 84 |
-
###
|
| 85 |
|
| 86 |
```bash
|
| 87 |
# Runtime stage and hardware
|
|
@@ -97,7 +106,23 @@ gradio predict build-small-hackathon/tiny-aya-compare /compare \
|
|
| 97 |
|
| 98 |
Expect `requested_hardware: zero-a10g` and both models loading at startup (~6 GB each in bf16).
|
| 99 |
|
| 100 |
-
###
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 101 |
|
| 102 |
| Change | Action |
|
| 103 |
|--------|--------|
|
|
|
|
| 48 |
hf spaces settings build-small-hackathon/tiny-aya-compare --hardware zero-a10g
|
| 49 |
```
|
| 50 |
|
| 51 |
+
### 2. Accept gated model access
|
| 52 |
|
| 53 |
+
Both checkpoints are **gated** (`auto`). The account whose token you store as `HF_TOKEN` must accept access first:
|
| 54 |
+
|
| 55 |
+
1. https://huggingface.co/CohereLabs/tiny-aya-global — click **Agree and access**
|
| 56 |
+
2. https://huggingface.co/CohereLabs/tiny-aya-earth — click **Agree and access**
|
| 57 |
+
|
| 58 |
+
Without this, startup fails with `403 Client Error` / `Cannot access gated repo`.
|
| 59 |
+
|
| 60 |
+
### 3. Add the gated-model token
|
| 61 |
+
|
| 62 |
+
Add a Space secret so the runtime can download the weights:
|
| 63 |
|
| 64 |
```bash
|
| 65 |
+
hf spaces secrets add build-small-hackathon/tiny-aya-compare -s HF_TOKEN
|
|
|
|
|
|
|
| 66 |
```
|
| 67 |
|
| 68 |
+
This uses your logged-in `hf` token. Re-run after accepting model access if the Space already failed once.
|
| 69 |
+
|
| 70 |
+
Use a token with `read` scope and access to both CohereLabs repos.
|
| 71 |
|
| 72 |
+
### 4. Push the app
|
| 73 |
|
| 74 |
From this repo:
|
| 75 |
|
|
|
|
| 90 |
--include "requirements.txt"
|
| 91 |
```
|
| 92 |
|
| 93 |
+
### 5. Verify the deployment
|
| 94 |
|
| 95 |
```bash
|
| 96 |
# Runtime stage and hardware
|
|
|
|
| 106 |
|
| 107 |
Expect `requested_hardware: zero-a10g` and both models loading at startup (~6 GB each in bf16).
|
| 108 |
|
| 109 |
+
### Troubleshooting
|
| 110 |
+
|
| 111 |
+
| Symptom | Fix |
|
| 112 |
+
|---------|-----|
|
| 113 |
+
| `Cannot access gated repo` / `403` on startup | Accept access on both model pages (step 2), then **Factory reboot** the Space |
|
| 114 |
+
| `RUNTIME_ERROR` after changing secrets | Factory reboot: Space Settings → Factory rebuild |
|
| 115 |
+
| `ZeroGPU quota exceeded` (visitors) | Visitors need GPU quota; creators don't pay per visit |
|
| 116 |
+
| Hardware shows `cpu-basic` | `hf spaces settings build-small-hackathon/tiny-aya-compare --hardware zero-a10g` |
|
| 117 |
+
|
| 118 |
+
Restart after fixing access:
|
| 119 |
+
|
| 120 |
+
```bash
|
| 121 |
+
# Re-upload or reboot from the Space Settings UI → "Factory rebuild"
|
| 122 |
+
hf spaces info build-small-hackathon/tiny-aya-compare --expand runtime
|
| 123 |
+
```
|
| 124 |
+
|
| 125 |
+
### 6. Update after changes
|
| 126 |
|
| 127 |
| Change | Action |
|
| 128 |
|--------|--------|
|