feat: abandon replicate client

Argcfile.sh

@@ -274,43 +274,6 @@ chat-vertexai() {
 -d "$(_build_body vertexai "$@")" 
 }
 
-# @cmd Chat with replicate api
-# @env REPLICATE_API_KEY!
-# @option -m --model=meta/meta-llama-3-8b-instruct $REPLICATE_MODEL
-# @flag -S --no-stream
-# @arg text~
-chat-replicate() {
-    url="https://api.replicate.com/v1/models/$argc_model/predictions"
-    res="$(_wrapper curl -s "$url" \
--X POST \
--H "Authorization: Bearer $REPLICATE_API_KEY" \
--H "Content-Type: application/json" \
--d "$(_build_body replicate "$@")" \
-)"
-    echo "$res"
-    if [[ -n "$argc_no_stream" ]]; then
-        prediction_url="$(echo "$res" | jq -r '.urls.get')"
-        while true; do
-            output="$(_wrapper curl -s -H "Authorization: Bearer $REPLICATE_API_KEY" "$prediction_url")"
-            prediction_status=$(printf "%s" "$output" | jq -r .status)
-            if [ "$prediction_status"=="succeeded" ]; then
-                echo "$output"
-                break
-            fi
-            if [ "$prediction_status"=="failed" ]; then
-                exit 1
-            fi
-            sleep 2
-        done
-    else
-        stream_url="$(echo "$res" | jq -r '.urls.stream')"
-    _wrapper curl -i --no-buffer "$stream_url" \
--H "Accept: text/event-stream" \
-
-    fi
-
-}
-
 # @cmd Chat with ernie api
 # @meta require-tools jq
 # @env ERNIE_API_KEY!
@@ -367,7 +330,7 @@ _choice_platform() {
 }
 
 _choice_client() {
-    printf "%s\n" openai gemini claude cohere ollama azure-openai vertexai bedrock cloudflare replicate ernie qianwen moonshot
+    printf "%s\n" openai gemini claude cohere ollama azure-openai vertexai bedrock cloudflare ernie qianwen moonshot
 }
 
 _choice_openai_compatible_platform() {
@@ -445,14 +408,6 @@ _build_body() {
         }
     ],
     "stream": '$stream'
-}'
-            ;;
-        replicate)
-            echo '{
-    "stream": '$stream',
-	"input": {
-      "prompt": "'"$*"'"
-	}
 }'
             ;;
         *)

README.md

@@ -50,7 +50,6 @@ Effortlessly connect with over 20 leading LLM platforms through a unified interf
 - **Perplexity:** Llama-3/Mixtral (paid, chat, online)
 - **Cloudflare:** (free, chat, embedding)
 - **OpenRouter:** (paid, chat, function-calling)
-- **Replicate:** (paid, chat)
 - **Ernie:** (paid, chat, embedding, reranker, function-calling)
 - **Qianwen:** Qwen (paid, chat, embedding, vision, function-calling)
 - **Moonshot:** (paid, chat, function-calling)

config.example.yaml

@@ -237,10 +237,6 @@ clients:
     api_base: https://api-inference.huggingface.co/v1
     api_key: xxx
 
-  # See https://replicate.com/docs
-  - type: replicate
-    api_key: xxx
-
   # See https://cloud.baidu.com/doc/WENXINWORKSHOP/index.html
   - type: ernie
     api_key: xxx

models.yaml

@@ -642,30 +642,6 @@
       input_price: 0
       output_price: 0
 
-# Links:
-#  - https://replicate.com/explore
-#  - https://replicate.com/pricing
-#  - https://replicate.com/docs/reference/http#create-a-prediction-using-an-official-model
-- platform: replicate
-  models:
-    - name: meta/meta-llama-3.1-405b-instruct
-      max_input_tokens: 128000
-      max_output_tokens: 4096
-      input_price: 9.5
-      output_price: 9.5
-    - name: meta/meta-llama-3-70b-instruct
-      max_input_tokens: 8192
-      max_output_tokens: 4096
-      require_max_tokens: true
-      input_price: 0.65
-      output_price: 2.75
-    - name: meta/meta-llama-3-8b-instruct
-      max_input_tokens: 8192
-      max_output_tokens: 4096
-      require_max_tokens: true
-      input_price: 0.05
-      output_price: 0.25
-
 # Links:
 #  - https://cloud.baidu.com/doc/WENXINWORKSHOP/s/Nlks5zkzu
 #  - https://cloud.baidu.com/doc/WENXINWORKSHOP/s/hlrk4akp7

src/client/mod.rs

@@ -4,7 +4,6 @@ mod message;
 #[macro_use]
 mod macros;
 mod model;
-mod prompt_format;
 mod stream;
 
 pub use crate::function::{ToolCall, ToolResults};
@@ -33,7 +32,6 @@ register_client!(
     ),
     (vertexai, "vertexai", VertexAIConfig, VertexAIClient),
     (bedrock, "bedrock", BedrockConfig, BedrockClient),
-    (replicate, "replicate", ReplicateConfig, ReplicateClient),
     (ernie, "ernie", ErnieConfig, ErnieClient),
 );