inst_llm 0.2.0 → 0.2.2

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 105c04ef5d12358f4ff663bc8c1278e731c0664a40ac016e29ec2defc999b6a3
-  data.tar.gz: 4873e34b6e2677822a6f8cabb3cac06367a6062d04301810856b33b751b260fb
+  metadata.gz: ca52be39233578c28b7d4b2e7a91b46c0b972d3806050ebb64cf7874cb11ca6c
+  data.tar.gz: ba51f3024ad3982515a265ed427c99e01c814069804e1e6a917f516a02999362
 SHA512:
-  metadata.gz: 6d7c85ea828972e507b8c680e340fb5ad76b5784277a441ecad7fb1ee215a2bc839c426daf76207cacf63b1ca9fe54ff4d7bf1bf3b9d1f78ef413a28e64ddc33
-  data.tar.gz: 7ffed97282a142a6adb4227334ddc333825b2de7225ec61b2f8d1796681a68b693be4799f73aee09cc4f5b482c2c8722a7633f9973d47bb42a8546880d75b997
+  metadata.gz: 2e713687d83aedbd2b6ab655f849dc29fb544fb5ea74b76f52547022646edba3e6ed4a1c9eaf45d44dc313d3cb39cfe7e1dafbd16483011a7fd7cba767899148
+  data.tar.gz: 2a8591d2ecd8019ad46e0d5adda72ddc652a2de3719f239ecc78dcac3f79482feccb97bbc86812cb673dd48c61b45bc23d0a296f60bfa38e188e83a9850c63ef

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 },
@@ -48,7 +51,10 @@ module InstLLM
       params = params_factory(model, messages, **options)
 
       begin
-        res = client.invoke_model(**params)
+        res = client.invoke_model(
+          content_type: "application/json",
+          **params
+        )
       rescue => error
         raise map_error_type(error)
       end
@@ -107,13 +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/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 = "0.2.0"
+  VERSION = "0.2.2"
 end

metadata

@@ -1,14 +1,14 @@
 --- !ruby/object:Gem::Specification
 name: inst_llm
 version: !ruby/object:Gem::Version
-  version: 0.2.0
+  version: 0.2.2
 platform: ruby
 authors:
 - Zach Pendleton
 autorequire:
 bindir: exe
 cert_chain: []
-date: 2024-04-18 00:00:00.000000000 Z
+date: 2024-04-25 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