Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
|
@@ -154,8 +154,8 @@ def submit(
|
|
| 154 |
"framework": framework_selector.lower(),
|
| 155 |
"image": {
|
| 156 |
"custom": {
|
| 157 |
-
health_route: "/health",
|
| 158 |
-
env: {
|
| 159 |
"DISABLE_CUSTOM_KERNELS": "true" if custom_kernel == "Enabled" else "false",
|
| 160 |
"MAX_BATCH_PREFILL_TOKENS": str(max_batch_prefill_token),
|
| 161 |
"MAX_BATCH_TOTAL_TOKENS": str(max_batch_total_token),
|
|
@@ -164,7 +164,7 @@ def submit(
|
|
| 164 |
"MODEL_ID": repository_selector.lower(),
|
| 165 |
# QUANTIZE: 'bitsandbytes' | 'gptq';
|
| 166 |
},
|
| 167 |
-
url: "ghcr.io/huggingface/text-generation-inference:1.0.1",
|
| 168 |
}
|
| 169 |
},
|
| 170 |
"repository": repository_selector.lower(),
|
|
|
|
| 154 |
"framework": framework_selector.lower(),
|
| 155 |
"image": {
|
| 156 |
"custom": {
|
| 157 |
+
"health_route": "/health",
|
| 158 |
+
"env": {
|
| 159 |
"DISABLE_CUSTOM_KERNELS": "true" if custom_kernel == "Enabled" else "false",
|
| 160 |
"MAX_BATCH_PREFILL_TOKENS": str(max_batch_prefill_token),
|
| 161 |
"MAX_BATCH_TOTAL_TOKENS": str(max_batch_total_token),
|
|
|
|
| 164 |
"MODEL_ID": repository_selector.lower(),
|
| 165 |
# QUANTIZE: 'bitsandbytes' | 'gptq';
|
| 166 |
},
|
| 167 |
+
"url": "ghcr.io/huggingface/text-generation-inference:1.0.1",
|
| 168 |
}
|
| 169 |
},
|
| 170 |
"repository": repository_selector.lower(),
|