inst_llm 0.2.1 → 0.2.3

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 9dd355405ae5b7659a6cee01536e9ca422a0658208e125303a6206e4e6129863
-  data.tar.gz: 8e0955a5bf08923385e093a7a6d739ff224695a345d52684d2c435675e3aa4c9
+  metadata.gz: c3ef21142135987ed0d03bd322ca86bd478c299a2c1430d5afea62c714b4552e
+  data.tar.gz: 188e6a291efc1de2485d01496b01a20fae10d72d79a3a90c5153df90dc057c8d
 SHA512:
-  metadata.gz: 787201e52ae44238a3a22d53a93f83fd15718b251a49b46fded527aab1e30eda9fe805574c1f42c0369a5586679a9d8f27c0cae5bb689c47c2c8fb65355e81b3
-  data.tar.gz: 14ea0088107fc9be654483530f314b5f00ea4d72686326b0d6c1349470446de8b8731fe06313081a1b78c78d6b9f2d8bff125752e60183b2dc3978626ffaf83a
+  metadata.gz: 904b2a084d3db540f620a030178a8f6141d3c02d87f942703850a234f7e2e0dccf7ee209abadfd57bc3f9c85c662889601a24e21ae9df59ae53f5aecde7fa7e8
+  data.tar.gz: 0bc2d34e8f9f19a836d5c96732990cba1cf824cd16722de7811ae73f2410c2ec0de738d0e96c86faf088f4ec8111bcb6564ac2a377ed539aac632bf7b785221d

data/lib/inst_llm/client.rb

@@ -12,6 +12,9 @@ module InstLLM
       "anthropic.claude-3-sonnet-20240229-v1:0": { format: :claude, provider: :bedrock, type: :chat },
       "anthropic.claude-3-haiku-20240307-v1:0": { format: :claude, provider: :bedrock, type: :chat },
 
+      "meta.llama3-8b-instruct-v1:0": { format: :llama3, provider: :bedrock, type: :chat },
+      "meta.llama3-70b-instruct-v1:0": { format: :llama3, provider: :bedrock, type: :chat },
+
       "mistral.mistral-7b-instruct-v0:2": { format: :mistral, provider: :bedrock, type: :chat },
       "mistral.mixtral-8x7b-instruct-v0:1": { format: :mistral, provider: :bedrock, type: :chat },
       "mistral.mistral-large-2402-v1:0": { format: :mistral, provider: :bedrock, type: :chat },
@@ -110,13 +113,14 @@ module InstLLM
       params_table = {
         claude: Parameter::ClaudeParameters,
         cohere_embed: Parameter::CohereEmbedParameters,
+        llama3: Parameter::Llama3Parameters,
         mistral: Parameter::MistralParameters
       }
       params_table[MODELS[model][:format]].new(model: model, messages: messages, **options)
     end
 
     def embedding_response_factory(model, response)
-      Response::EmbeddingResponse.send(:"from_#{MODELS[model][:format]}",  model: model, response: response)
+      Response::EmbeddingResponse.send(:"from_#{MODELS[model][:format]}", model: model, response: response)
     end
 
     def response_factory(model, response)

data/lib/inst_llm/parameter/all.rb

@@ -1,5 +1,6 @@
 # frozen_string_literal: true
 
-require_relative 'claude_parameters'
-require_relative 'cohere_embed_parameters'
-require_relative 'mistral_parameters'
+require_relative "claude_parameters"
+require_relative "cohere_embed_parameters"
+require_relative "llama3_parameters"
+require_relative "mistral_parameters"

data/lib/inst_llm/parameter/llama3_parameters.rb

@@ -0,0 +1,60 @@
+# frozen_string_literal: true
+
+require "json"
+
+module InstLLM
+  module Parameter
+    class Llama3Parameters
+      DEFAULT_OPTIONS = {
+        max_gen_len: nil,
+        temperature: nil,
+        top_p: nil,
+      }.freeze
+
+      def initialize(model:, messages: [], **options)
+        @model = model
+        @messages = messages.map { |message| format_message(message) }
+        @options = DEFAULT_OPTIONS.merge(options.slice(*DEFAULT_OPTIONS.keys)).compact
+      end
+
+      def to_hash
+        { model_id: @model, body: { prompt: prompt(@messages) }.merge(@options).to_json }
+      end
+
+      private
+
+      def format_message(message)
+        if message[:content].is_a?(String)
+          message
+        elsif message[:content].is_a?(Array)
+          content = message[:content].map do |m|
+            raise UnknownArgumentError, <<~ERR unless m[:type].to_sym == :text
+              Model does not support multiple types. Expected :text, received #{m[:type]}"
+            ERR
+
+            m[:text]
+          end
+          { role: message[:role], content: content.join("\n\n") }
+        else
+          raise ArgumentError, "Message content must be a String or Array; received: #{message[:content]}"
+        end
+      end
+
+      def prompt(messages)
+        messages << { role: :assistant, content: "" } if messages.last[:role].to_sym != :assistant
+
+        formatted_messages = messages.map do |message|
+          "<|start_header_id|>#{message[:role].to_s}<|end_header_id|>\n#{message[:content]}"
+        end
+
+        formatted_prompt = "<|begin_of_text|>#{formatted_messages.join("<|eot_id|>")}"
+
+        if messages.last[:role] == :user
+          formatted_prompt += "<|eot_id|>"
+        end
+
+        formatted_prompt
+      end
+    end
+  end
+end

data/lib/inst_llm/parameter/mistral_parameters.rb

@@ -23,9 +23,23 @@ module InstLLM
 
       private
 
+      def format_message(message)
+        return message if message[:content].is_a?(String)
+
+        content = Array(message[:content]).map do |m|
+          raise UnknownArgumentError, <<~ERR if m[:type] != :text
+            Model does not support multiple types. Expected :text, received #{m[:type]} 
+          ERR
+
+          m[:text]
+        end
+
+        { role: message[:role], content: content.join("\n\n") }
+      end
+
       def prompt
         system_message = nil
-        prompt = @messages.map do |message|
+        prompt = @messages.map { |message| format_message(message) }.map do |message|
           case message[:role].to_sym
           when :assistant
             "#{message[:content]}"

data/lib/inst_llm/response/chat_response.rb

@@ -29,6 +29,18 @@ module InstLLM
           )
         end
 
+        def from_llama3(model:, response:)
+          new(
+            model: model,
+            message: { role: :assistant, content: response["generation"] },
+            stop_reason: response["stop_reason"],
+            usage: {
+              input_tokens: response["prompt_token_count"],
+              output_tokens: response["generation_token_count"]
+            }
+          )
+        end
+
         def from_mistral(model:, response:)
           new(
             model: model,

data/lib/inst_llm/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module InstLLM
-  VERSION = "0.2.1"
+  VERSION = "0.2.3"
 end

metadata

@@ -1,14 +1,14 @@
 --- !ruby/object:Gem::Specification
 name: inst_llm
 version: !ruby/object:Gem::Version
-  version: 0.2.1
+  version: 0.2.3
 platform: ruby
 authors:
 - Zach Pendleton
 autorequire:
 bindir: exe
 cert_chain: []
-date: 2024-04-23 00:00:00.000000000 Z
+date: 2024-04-26 00:00:00.000000000 Z
 dependencies:
 - !ruby/object:Gem::Dependency
   name: aws-sdk-bedrockruntime
@@ -36,6 +36,7 @@ files:
 - lib/inst_llm/parameter/all.rb
 - lib/inst_llm/parameter/claude_parameters.rb
 - lib/inst_llm/parameter/cohere_embed_parameters.rb
+- lib/inst_llm/parameter/llama3_parameters.rb
 - lib/inst_llm/parameter/mistral_parameters.rb
 - lib/inst_llm/response/all.rb
 - lib/inst_llm/response/chat_response.rb