From a732291f3321f2feb31fb7b1105e223e2fa33976 Mon Sep 17 00:00:00 2001 From: sigoden Date: Wed, 5 Jun 2024 09:10:12 +0800 Subject: refactor: rename `pass_max_tokens` to `require_max_tokens` (#562) --- assets/arena.html | 8 +++---- assets/playground.html | 8 +++---- models.yaml | 64 +++++++++++++++++++++++++------------------------- src/client/model.rs | 8 +++---- src/serve.rs | 4 ++-- 5 files changed, 46 insertions(+), 46 deletions(-) diff --git a/assets/arena.html b/assets/arena.html index 148ec4b..b43c464 100644 --- a/assets/arena.html +++ b/assets/arena.html @@ -824,8 +824,8 @@ messages: messages, stream: true, }; - const { max_output_token, pass_max_tokens } = retrieveModel(this.models, chat.model_id); - if (!body["max_tokens"] && pass_max_tokens) { + const { max_output_token, require_max_tokens } = retrieveModel(this.models, chat.model_id); + if (!body["max_tokens"] && require_max_tokens) { body["max_tokens"] = max_output_token; }; return body; @@ -907,12 +907,12 @@ if (!model) return {}; const max_output_token = model.max_output_tokens; const supports_vision = !!model.supports_vision; - const pass_max_tokens = !!model.pass_max_tokens; + const require_max_tokens = !!model.require_max_tokens; return { id, max_output_token, supports_vision, - pass_max_tokens, + require_max_tokens, } } diff --git a/assets/playground.html b/assets/playground.html index 285dbae..4487b44 100644 --- a/assets/playground.html +++ b/assets/playground.html @@ -1032,8 +1032,8 @@ body[body_key || setting_key] = this.settings[setting_key]; } }); - const { max_output_token, pass_max_tokens } = this.currentModel; - if (!body["max_tokens"] && pass_max_tokens) { + const { max_output_token, require_max_tokens } = this.currentModel; + if (!body["max_tokens"] && require_max_tokens) { body["max_tokens"] = max_output_token; }; return body; @@ -1116,12 +1116,12 @@ if (!model) return {}; const max_output_token = model.max_output_tokens; const supports_vision = !!model.supports_vision; - const pass_max_tokens = !!model.pass_max_tokens; + const require_max_tokens = !!model.require_max_tokens; return { id, max_output_token, supports_vision, - pass_max_tokens, + require_max_tokens, } } diff --git a/models.yaml b/models.yaml index f4c980b..c0ac664 100644 --- a/models.yaml +++ b/models.yaml @@ -51,7 +51,7 @@ - name: gpt-4-vision-preview max_input_tokens: 128000 max_output_tokens: 4096 - pass_max_tokens: true + require_max_tokens: true input_price: 10 output_price: 30 supports_vision: true @@ -125,7 +125,7 @@ - name: claude-3-opus-20240229 max_input_tokens: 200000 max_output_tokens: 4096 - pass_max_tokens: true + require_max_tokens: true input_price: 15 output_price: 75 supports_vision: true @@ -133,7 +133,7 @@ - name: claude-3-sonnet-20240229 max_input_tokens: 200000 max_output_tokens: 4096 - pass_max_tokens: true + require_max_tokens: true input_price: 3 output_price: 15 supports_vision: true @@ -141,7 +141,7 @@ - name: claude-3-haiku-20240307 max_input_tokens: 200000 max_output_tokens: 4096 - pass_max_tokens: true + require_max_tokens: true input_price: 0.25 output_price: 1.25 supports_vision: true @@ -330,21 +330,21 @@ - name: claude-3-opus@20240229 max_input_tokens: 200000 max_output_tokens: 4096 - pass_max_tokens: true + require_max_tokens: true input_price: 15 output_price: 75 supports_vision: true - name: claude-3-sonnet@20240229 max_input_tokens: 200000 max_output_tokens: 4096 - pass_max_tokens: true + require_max_tokens: true input_price: 3 output_price: 15 supports_vision: true - name: claude-3-haiku@20240307 max_input_tokens: 200000 max_output_tokens: 4096 - pass_max_tokens: true + require_max_tokens: true input_price: 0.25 output_price: 1.25 supports_vision: true @@ -360,52 +360,52 @@ - name: anthropic.claude-3-opus-20240229-v1:0 max_input_tokens: 200000 max_output_tokens: 4096 - pass_max_tokens: true + require_max_tokens: true input_price: 15 output_price: 75 supports_vision: true - name: anthropic.claude-3-sonnet-20240229-v1:0 max_input_tokens: 200000 max_output_tokens: 4096 - pass_max_tokens: true + require_max_tokens: true input_price: 3 output_price: 15 supports_vision: true - name: anthropic.claude-3-haiku-20240307-v1:0 max_input_tokens: 200000 max_output_tokens: 4096 - pass_max_tokens: true + require_max_tokens: true input_price: 0.25 output_price: 1.25 supports_vision: true - name: meta.llama3-8b-instruct-v1:0 max_input_tokens: 8192 max_output_tokens: 2048 - pass_max_tokens: true + require_max_tokens: true input_price: 0.4 output_price: 0.6 - name: meta.llama3-70b-instruct-v1:0 max_input_tokens: 8192 max_output_tokens: 2048 - pass_max_tokens: true + require_max_tokens: true input_price: 2.65 output_price: 3.5 - name: mistral.mistral-7b-instruct-v0:2 max_input_tokens: 32000 max_output_tokens: 8192 - pass_max_tokens: true + require_max_tokens: true input_price: 0.15 output_price: 0.2 - name: mistral.mixtral-8x7b-instruct-v0:1 max_input_tokens: 32000 max_output_tokens: 8192 - pass_max_tokens: true + require_max_tokens: true input_price: 0.45 output_price: 0.7 - name: mistral.mistral-large-2402-v1:0 max_input_tokens: 32000 max_output_tokens: 8192 - pass_max_tokens: true + require_max_tokens: true input_price: 8 output_price: 2.4 @@ -419,23 +419,23 @@ - name: '@cf/meta/llama-3-8b-instruct' max_input_tokens: 4096 max_output_tokens: 4096 - pass_max_tokens: true + require_max_tokens: true - name: '@cf/mistral/mistral-7b-instruct-v0.2-lora' max_input_tokens: 4096 max_output_tokens: 4096 - pass_max_tokens: true + require_max_tokens: true - name: '@cf/google/gemma-7b-it-lora' max_input_tokens: 4096 max_output_tokens: 4096 - pass_max_tokens: true + require_max_tokens: true - name: '@cf/qwen/qwen1.5-14b-chat-awq' max_input_tokens: 4096 max_output_tokens: 4096 - pass_max_tokens: true + require_max_tokens: true - name: '@hf/thebloke/deepseek-coder-6.7b-instruct-awq' max_input_tokens: 4096 max_output_tokens: 4096 - pass_max_tokens: true + require_max_tokens: true - platform: replicate # docs: @@ -448,25 +448,25 @@ - name: meta/meta-llama-3-70b-instruct max_input_tokens: 8192 max_output_tokens: 4096 - pass_max_tokens: true + 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 - pass_max_tokens: true + require_max_tokens: true input_price: 0.05 output_price: 0.25 - name: mistralai/mistral-7b-instruct-v0.2 max_input_tokens: 32000 max_output_tokens: 8192 - pass_max_tokens: true + require_max_tokens: true input_price: 0.05 output_price: 0.25 - name: mistralai/mixtral-8x7b-instruct-v0.1 max_input_tokens: 32000 max_output_tokens: 8192 - pass_max_tokens: true + require_max_tokens: true input_price: 0.3 output_price: 1 @@ -480,31 +480,31 @@ - name: ernie-4.0-8k-preview max_input_tokens: 5120 max_output_tokens: 2048 - pass_max_tokens: true + require_max_tokens: true input_price: 16.8 output_price: 16.8 - name: ernie-3.5-8k-preview max_input_tokens: 5120 max_output_tokens: 2048 - pass_max_tokens: true + require_max_tokens: true input_price: 1.68 output_price: 1.68 - name: ernie-speed-128k max_input_tokens: 124000 max_output_tokens: 4096 - pass_max_tokens: true + require_max_tokens: true input_price: 0 output_price: 0 - name: ernie-lite-8k max_input_tokens: 7168 max_output_tokens: 2048 - pass_max_tokens: true + require_max_tokens: true input_price: 0 output_price: 0 - name: ernie-tiny-8k max_input_tokens: 7168 max_output_tokens: 2048 - pass_max_tokens: true + require_max_tokens: true input_price: 0 output_price: 0 @@ -810,21 +810,21 @@ - name: anthropic/claude-3-opus max_input_tokens: 200000 max_output_tokens: 4096 - pass_max_tokens: true + require_max_tokens: true input_price: 15 output_price: 75 supports_vision: true - name: anthropic/claude-3-sonnet max_input_tokens: 200000 max_output_tokens: 4096 - pass_max_tokens: true + require_max_tokens: true input_price: 3 output_price: 15 supports_vision: true - name: anthropic/claude-3-haiku max_input_tokens: 200000 max_output_tokens: 4096 - pass_max_tokens: true + require_max_tokens: true input_price: 0.25 output_price: 1.25 supports_vision: true diff --git a/src/client/model.rs b/src/client/model.rs index e16cb4e..22ad4d9 100644 --- a/src/client/model.rs +++ b/src/client/model.rs @@ -153,7 +153,7 @@ impl Model { } pub fn max_tokens_param(&self) -> Option { - if self.data.pass_max_tokens { + if self.data.require_max_tokens { self.data.max_output_tokens } else { None @@ -163,13 +163,13 @@ impl Model { pub fn set_max_tokens( &mut self, max_output_tokens: Option, - pass_max_tokens: bool, + require_max_tokens: bool, ) -> &mut Self { match max_output_tokens { None | Some(0) => self.data.max_output_tokens = None, _ => self.data.max_output_tokens = max_output_tokens, } - self.data.pass_max_tokens = pass_max_tokens; + self.data.require_max_tokens = require_max_tokens; self } @@ -227,7 +227,7 @@ pub struct ModelData { // chat-only properties pub max_output_tokens: Option, #[serde(default)] - pub pass_max_tokens: bool, + pub require_max_tokens: bool, #[serde(default)] pub supports_vision: bool, #[serde(default)] diff --git a/src/serve.rs b/src/serve.rs index 116942a..3f33fc2 100644 --- a/src/serve.rs +++ b/src/serve.rs @@ -85,7 +85,7 @@ impl Server { let ModelData { max_input_tokens, max_output_tokens, - pass_max_tokens, + require_max_tokens, input_price, output_price, supports_vision, @@ -96,7 +96,7 @@ impl Server { "id": id, "max_input_tokens": max_input_tokens, "max_output_tokens": max_output_tokens, - "pass_max_tokens": pass_max_tokens, + "require_max_tokens": require_max_tokens, "input_price": input_price, "output_price": output_price, "supports_vision": supports_vision, -- cgit v1.2.3