DeepInfra
/

Llama-2-70b-chat-hf-trt-fp8

Model card Files Files and versions Community

yessenzhar commited on Dec 12, 2023

Commit

5de4d5c

•

1 Parent(s): d5c7a29

remove hardcoded weights and tokenizer dirs, replace with template

Files changed (3) hide show

postprocessing/config.pbtxt +1 -1
preprocessing/config.pbtxt +1 -1
tensorrt_llm/config.pbtxt +1 -1

postprocessing/config.pbtxt CHANGED Viewed

@@ -50,7 +50,7 @@ output [
 parameters {
   key: "tokenizer_dir"
   value: {
-    string_value: "/data/tgi-data/orig_llama"
   }
 }

 parameters {
   key: "tokenizer_dir"
   value: {
+    string_value: "${tokenizer_dir}"
   }
 }

preprocessing/config.pbtxt CHANGED Viewed

@@ -80,7 +80,7 @@ output [
 parameters {
   key: "tokenizer_dir"
   value: {
-    string_value: "/data/tgi-data/orig_llama"
   }
 }

 parameters {
   key: "tokenizer_dir"
   value: {
+    string_value: "${tokenizer_dir}"
   }
 }

tensorrt_llm/config.pbtxt CHANGED Viewed

@@ -173,7 +173,7 @@ parameters: {
 parameters: {
   key: "gpt_model_path"
   value: {
-    string_value: "/data/tgi-data/triton_model_repo_70_fp8/tensorrt_llm/1"
   }
 }
 parameters: {

 parameters: {
   key: "gpt_model_path"
   value: {
+    string_value: "${gpt_model_path}"
   }
 }
 parameters: {