GGUF
Inference Endpoints
van-qa commited on
Commit
a18dea7
1 Parent(s): 062e706

Update metadata.yaml

Browse files
Files changed (1) hide show
  1. metadata.yaml +8 -7
metadata.yaml CHANGED
@@ -1,11 +1,12 @@
1
  name: mistral
2
 
3
  default:
4
- - main: Mistral-7B-Instruct-v0.3 Q4_K_M
5
- - gguf: Mistral-7B-Instruct-v0.3 Q4_K_M
6
- - onnx: Mistral-7B-Instruct-v0.3 AWQ INT4
7
- - tensorrt-llm: Mistral-7B-Instruct-v0.3 AWQ INT4
 
8
  7b:
9
- - gguf: Mistral-7B-Instruct-v0.3 Q4_K_M
10
- - onnx: Mistral-7B-Instruct-v0.3 AWQ INT4
11
- - tensorrt-llm: Mistral-7B-Instruct-v0.3 AWQ INT4
 
1
  name: mistral
2
 
3
  default:
4
+ - main: Q4_K_M
5
+ - gguf: Q4_K_M
6
+ - onnx: AWQ INT4
7
+ - tensorrt-llm: AWQ INT4
8
+
9
  7b:
10
+ - gguf: Q4_K_M
11
+ - onnx: AWQ INT4
12
+ - tensorrt-llm: AWQ INT4