ollama-ruby 0.11.0 → 0.12.0

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 6611c14b779a919b256552774080305825b4e81cd5e43ae0303026aaaed7c13e
-  data.tar.gz: a34e71ef5b07cdd8a0cdf5dce9de7f7e9c3faca1829e4837c54ea243458b8f05
+  metadata.gz: 417e37f903a49bd6cea779cbf15674cc89ebee9c9ea9bf8930fc2b1fb8007794
+  data.tar.gz: f171af3e86996daf776df3f2b28dbf39c71c23139118a879b36e9cbabae5f376
 SHA512:
-  metadata.gz: aa38db4bdd42ebffe2ef9b5abf7f644e38821acc4fb9650b7e8646a9fab3350640224c4ba59e13d7f182becd2a84ac881fd0c5d136ca88782e1a7febec31c2fe
-  data.tar.gz: b14f2626d9f2c8e6e8bc6c3b3d8947d78b4400cf0dd8888cf9f818261ba3605fc9971c2c9015f976ec48f2004cb4b0ff5d922e2efe001b5b443f03728f14ce51
+  metadata.gz: 7f236cf84e27feb750836fe695cf5812f4ac4af3fb63c6694e71ce6a077383abc14030aed184d668de59e9d082ba8ab890eddfc2052d817b27b708b69af540b9
+  data.tar.gz: 585818aceb2af7d0cad69566d6bd62c8b443650b03f844650831a8e6fcdfcd370b93f8127b85726548ff2fe08a817d5c4f1333c8f8af81af5d8777e8cc57610b

data/CHANGES.md

@@ -1,5 +1,43 @@
 # Changes
 
+## 2024-11-26 v0.12.0
+
+* **Upgrade display/clear links used in chat**:
+  * Created `$links` set to store used links.
+  * Added `/links` command to display used links as a enumerated list.
+  * Implemented `/links (clear)` feature to remove all or specific used links.
+* **Update semantic splitter to handle embeddings size < 2**:
+  + Added condition to return sentences directly when embeddings size is less
+    than 2.
+* **Removed collection list from chat info output**
+* **Add SQLiteCache spec for convert_to_vector method**:
+  - Test creates a vector with two elements and checks if
+    `cache.convert_to_vector(vector)` returns the same vector (which for this
+    cache is just a Ruby array).
+* **Add tests for retrieving tags from cache**:
+  * Test if tags are returned as an instance of `Ollama::Utils::Tags`
+  * Test also checks if the order of the tags is correct
+* **Added test case for clearing tags from `Ollama::Documents::SQLiteCache`**
+  - Updated spec for new `clear_for_tags` method
+* **Migrate SQLite cache to use new clear_for_tags method**:
+  + Added `clear_for_tags` method to SQLiteCache class in `sqlite_cache.rb`
+  + Modified `clear` method in `records.rb` to call `clear_for_tags` if
+    available
+  + Created `find_records_for_tags` method in `sqlite_cache.rb` to find records
+    by tags
+  + Updated `find_records` method in `sqlite_cache.rb` to use new
+    `find_records_for_tags` method
+* **Use Ollama::Utils::Tags for consistently handling tags**
+* **Upgrade SQLite cache to use correct prefix for full_each**:
+  * Use `?%` as the default prefix in `SQLiteCache#full_each`
+  * Add specs for setting keys with different prefixes in `SQLiteCache`
+  * Add specs for setting keys with different prefixes in `MemoryCache`
+* **Refactor SQLite cache query explanation**
+  + Use new variable `e` to store sanitized query for debugging purposes
+  + Pass sanitized query `e` to `@database.execute` for `EXPLAIN` instead of
+    original query `a[0]`
+* **Add test for unique tags with leading # characters**
+
 ## 2024-11-20 v0.11.0
 
 * Added `voice` and `interactive` reader attributes to the Say handler class.
@@ -25,7 +63,7 @@
   + Added support for `file://` protocol to content scans.
   + Updated regex pattern to match local files starting with `~`, `.`, or `/`.
   + Remove # anchors for file URLs (and files)
-* Improved parsing of content in ollama_chat:
+* Improved parsing of content in `ollama_chat`:
   + Use `content.scan(%r((https?://\S+)|(#\S+)|(\S+\/\S+)))` to match URLs, tags and files.
   + For foo/bar file pathes prepend `./`foo/bar, for foo you have to enter ./foo still.
   + Added a check for file existence before fetching its content

data/README.md

@@ -44,8 +44,8 @@ print ollama.chat(model: 'llama3.1', stream: true, messages:).lazy.map { |respon
 
 ## Try out things in ollama\_console
 
-This is an interactive console, that can be used to try the different commands
-provided by an `Ollama::Client` instance. For example this command generate a
+This is an interactive console where you can try out the different commands
+provided by an `Ollama::Client` instance. For example, this command generates a
 response and displays it on the screen using the Markdown handler:
 
 ```
@@ -459,6 +459,7 @@ The following commands can be given inside the chat, if prefixed by a `/`:
 /embedding                      toggle embedding paused or not
 /embed source                   embed the source's content
 /web [n] query                  query web search & return n or 1 results
+/links( clear)                  display (or clear) links used in the chat
 /save filename                  store conversation messages
 /load filename                  load conversation messages
 /quit                           to quit

data/Rakefile

@@ -42,7 +42,9 @@ GemHadar do
   dependency             'tins',                  '~> 1.34'
   dependency             'kramdown-ansi',         '~> 0.0', '>= 0.0.1'
   dependency             'ostruct',               '~> 0.0'
-  development_dependency 'all_images',            '~> 0.4'
+  dependency             'sqlite-vec',            '~> 0.0'
+  dependency             'sqlite3',               '~> 2.0', '>= 2.0.1'
+  development_dependency 'all_images',            '~> 0.6'
   development_dependency 'rspec',                 '~> 3.2'
   development_dependency 'webmock'
   development_dependency 'debug'

data/bin/ollama_chat

@@ -57,13 +57,15 @@ class OllamaChatConfig
       enabled: true
       model:
         name: mxbai-embed-large
+        embedding_length: 1024
         options: {}
         # Retrieval prompt template:
         prompt: 'Represent this sentence for searching relevant passages: %s'
       batch_size: 10
+      database_filename: null # ':memory:'
       collection: <%= ENV['OLLAMA_CHAT_COLLECTION'] %>
       found_texts_size: 4096
-      found_texts_count: null
+      found_texts_count: 10
       splitter:
         name: RecursiveCharacter
         chunk_size: 1024
@@ -81,12 +83,15 @@ class OllamaChatConfig
 
   def initialize(filename = nil)
     @filename = filename || default_path
+    unless File.directory?(cache_dir_path)
+      mkdir_p cache_dir_path.to_s
+    end
     @config = Provider.config(@filename, '⚙️')
     retried = false
   rescue ConfigurationFileMissing
     if @filename == default_path && !retried
       retried = true
-      mkdir_p File.dirname(default_path)
+      mkdir_p config_dir_path.to_s
       File.secure_write(default_path, DEFAULT_CONFIG)
       retry
     else
@@ -105,6 +110,14 @@ class OllamaChatConfig
   def config_dir_path
     XDG.new.config_home + 'ollama_chat'
   end
+
+  def cache_dir_path
+    XDG.new.cache_home + 'ollama_chat'
+  end
+
+  def database_path
+    cache_dir_path + 'documents.db'
+  end
 end
 
 class FollowChat
@@ -316,11 +329,12 @@ def search_web(query, n = nil)
     doc.css('.results_links').each do |link|
       if n > 0
         url = link.css('.result__a').first&.[]('href')
-        url.sub!(%r(\A/l/\?uddg=), '')
+        url.sub!(%r(\A(//duckduckgo\.com)?/l/\?uddg=), '')
         url.sub!(%r(&rut=.*), '')
         url = URI.decode_uri_component(url)
         url = URI.parse(url)
         url.host =~ /duckduckgo\.com/ and next
+        $links.add(url.to_s)
         result << url
         n -= 1
       else
@@ -387,7 +401,7 @@ module SourceParsing
       end
       source_io.rewind
       source_io.read
-    when 'text/csv' # TODO
+    when 'text/csv'
       parse_csv(source_io)
     when 'application/rss+xml'
       parse_rss(source_io)
@@ -510,6 +524,7 @@ def fetch_source(source, &block)
       block.(tmp)
     end
   when %r(\Ahttps?://\S+)
+    $links.add(source.to_s)
     Utils::Fetcher.get(
       source,
       cache:        $cache,
@@ -603,7 +618,6 @@ def embed_source(source_io, source, count: nil)
       percentage: splitter_config.percentage?,
       percentile: splitter_config.percentile?,
     )
-    inputs = splitter.split(text)
   end
   inputs or return
   source = source.to_s
@@ -624,7 +638,6 @@ def embed(source)
       content.present? or return
       source_io.rewind
       embed_source(source_io, source)
-      content
     end
     $config.prompts.embed % { source: }
   else
@@ -649,6 +662,7 @@ def parse_content(content, images)
       File.exist?(file) or next
       source = file
     when url
+      $links.add(url.to_s)
       source = url
     end
     fetch_source(source) do |source_io|
@@ -674,7 +688,7 @@ def parse_content(content, images)
       end
     end
   end
-  new_content = contents.select(&:present?).compact * "\n\n"
+  new_content = contents.select { _1.present? rescue nil }.compact * "\n\n"
   return new_content, (tags unless tags.empty?)
 end
 
@@ -735,17 +749,12 @@ ensure
 end
 
 def collection_stats
-  list = $documents.collections.sort.map { |c|
-    '  ' + ($documents.collection == c ? bold { c } : c).to_s
-  }.join(?\n)
   puts <<~EOT
     Current Collection
       Name: #{bold{$documents.collection}}
       #Embeddings: #{$documents.size}
       #Tags: #{$documents.tags.size}
       Tags: #{$documents.tags}
-    List:
-    #{list}
   EOT
 end
 
@@ -785,10 +794,11 @@ def set_system_prompt(messages, system)
   messages << Message.new(role: 'system', content: system)
 end
 
-def change_system_prompt(messages, default)
-  prompts = $config.system_prompts.attribute_names.compact
-  chosen  = Utils::Chooser.choose(prompts)
-  system  = if chosen
+def change_system_prompt(messages, default, system: nil)
+  selector = Regexp.new(system.to_s[1..-1].to_s)
+  prompts  = $config.system_prompts.attribute_names.compact.grep(selector)
+  chosen   = Utils::Chooser.choose(prompts, return_immediately: true)
+  system   = if chosen
               $config.system_prompts.send(chosen)
             else
               default
@@ -869,6 +879,7 @@ def display_chat_help
     /embedding                      toggle embedding paused or not
     /embed source                   embed the source's content
     /web [n] query                  query web search & return n or 1 results
+    /links( clear)                  display (or clear) links used in the chat
     /save filename                  store conversation messages
     /load filename                  load conversation messages
     /quit                           to quit
@@ -907,16 +918,17 @@ end
 
 $opts = go 'f:u:m:s:c:C:D:MEVh'
 
-config = OllamaChatConfig.new($opts[?f])
-$config = config.config
+$ollama_chat_config = OllamaChatConfig.new($opts[?f])
+$config             = $ollama_chat_config.config
 
 setup_switches
 
 $opts[?h] and usage
 $opts[?V] and version
 
-base_url = $opts[?u] || $config.url
-$ollama  = Client.new(base_url:, debug: $config.debug)
+base_url   = $opts[?u] || $config.url
+user_agent = [ File.basename($0), Ollama::VERSION ] * ?/
+$ollama    = Client.new(base_url:, debug: $config.debug, user_agent:)
 
 $document_policy = $config.document_policy
 $model           = choose_model($opts[?m], $config.model.name)
@@ -929,8 +941,8 @@ if $opts[?c]
   messages.concat load_conversation($opts[?c])
 else
   default = $config.system_prompts.default? || model_system
-  if $opts[?s] == ??
-    change_system_prompt(messages, default)
+  if $opts[?s] =~ /\A\?/
+    change_system_prompt(messages, default, system: $opts[?s])
   else
     system = Utils::FileArgument.get_file_argument($opts[?s], default:)
     system.present? and set_system_prompt(messages, system)
@@ -944,12 +956,13 @@ if $embedding.on?
   collection = $opts[?C] || $config.embedding.collection
   $documents = Documents.new(
     ollama:,
-    model:         $embedding_model,
-    model_options: $config.embedding.model.options,
-    collection:,
-    cache:         configure_cache,
-    redis_url:     $config.redis.documents.url?,
-    debug:         ENV['DEBUG'].to_i == 1,
+    model:             $embedding_model,
+    model_options:     $config.embedding.model.options,
+    database_filename: $config.embedding.database_filename || $ollama_chat_config.database_path,
+    collection:        ,
+    cache:             configure_cache,
+    redis_url:         $config.redis.documents.url?,
+    debug:             $config.debug
   )
 
   document_list = $opts[?D].to_a
@@ -991,10 +1004,11 @@ end
 
 $current_voice = $config.voice.default
 
-puts "Configuration read from #{config.filename.inspect} is:", $config
+puts "Configuration read from #{$ollama_chat_config.filename.inspect} is:", $config
 info
 puts "\nType /help to display the chat help."
 
+$links  = Set.new
 images = []
 loop do
   parse_content = true
@@ -1134,6 +1148,37 @@ loop do
     save_conversation($1, messages)
     puts "Saved conversation to #$1."
     next
+  when %r(^/links(?:\s+(clear))?$)
+    case $1
+    when 'clear'
+      loop do
+        links = $links.dup.add('[EXIT]').add('[ALL]')
+        link = Utils::Chooser.choose(links, prompt: 'Clear? %s')
+        case link
+        when nil, '[EXIT]'
+          puts "Exiting chooser."
+          break
+        when '[ALL]'
+          if ask?(prompt: 'Are you sure? (y/n) ') =~ /\Ay/i
+            $links.clear
+            puts "Cleared all links in list."
+            break
+          else
+            puts 'Cancelled.'
+            sleep 3
+          end
+        when /./
+          $links.delete(link)
+          puts "Cleared link from links in list."
+          sleep 3
+        end
+      end
+    when nil
+      format  = "% #{Math.log10($links.size).ceil}s. %s"
+      connect = -> link { hyperlink(link) { link } }
+      puts $links.each_with_index.map { |x, i| format % [ i + 1, connect.(x) ] }
+    end
+    next
   when %r(^/load\s+(.+)$)
     messages = load_conversation($1)
     puts "Loaded conversation from #$1."

data/lib/ollama/client.rb

@@ -10,7 +10,7 @@ class Ollama::Client
 
   annotate :doc
 
-  def initialize(base_url: nil, output: $stdout, connect_timeout: nil, read_timeout: nil, write_timeout: nil, debug: nil)
+  def initialize(base_url: nil, output: $stdout, connect_timeout: nil, read_timeout: nil, write_timeout: nil, debug: nil, user_agent: nil)
     base_url.nil? and base_url = ENV.fetch('OLLAMA_URL') do
       raise ArgumentError,
         'missing :base_url parameter or OLLAMA_URL environment variable'
@@ -21,8 +21,8 @@ class Ollama::Client
     @ssl_verify_peer = base_url.query.to_s.split(?&).inject({}) { |h, l|
       h.merge Hash[*l.split(?=)]
     }['ssl_verify_peer'] != 'false'
-    @base_url, @output, @connect_timeout, @read_timeout, @write_timeout, @debug =
-      base_url, output, connect_timeout, read_timeout, write_timeout, debug
+    @base_url, @output, @connect_timeout, @read_timeout, @write_timeout, @debug, @user_agent =
+      base_url, output, connect_timeout, read_timeout, write_timeout, debug, user_agent
   end
 
   attr_accessor :output
@@ -111,13 +111,13 @@ class Ollama::Client
 
   def headers
     {
-      'User-Agent'   => self.class.user_agent,
+      'User-Agent'   => @user_agent || self.class.user_agent,
       'Content-Type' => 'application/json; charset=utf-8',
     }
   end
 
   def self.user_agent
-    '%s/%s' % [ self.class, Ollama::VERSION ]
+    '%s/%s' % [ self, Ollama::VERSION ]
   end
 
   def excon(url)

data/lib/ollama/documents/cache/common.rb

@@ -1,7 +1,11 @@
 module Ollama::Documents::Cache::Common
   include Ollama::Utils::Math
 
-  attr_writer :prefix # current prefix defined for the cache
+  def initialize(prefix:)
+    self.prefix = prefix
+  end
+
+  attr_accessor :prefix # current prefix defined for the cache
 
   # Returns an array of collection names that match the given prefix.
   #

data/lib/ollama/documents/cache/memory_cache.rb

@@ -4,7 +4,7 @@ class Ollama::Documents::MemoryCache
   include Ollama::Documents::Cache::Common
 
   def initialize(prefix:)
-    @prefix = prefix
+    super(prefix:)
     @data   = {}
   end
 

data/lib/ollama/documents/cache/records.rb

@@ -0,0 +1,87 @@
+module Ollama::Documents::Cache::Records
+  class Record < JSON::GenericObject
+    def initialize(*a)
+      super
+      self.text ||= ''
+      self.norm ||= 0.0
+    end
+
+    def to_s
+      my_tags = tags_set
+      my_tags.empty? or my_tags = " #{my_tags}"
+      "#<#{self.class} #{text.inspect}#{my_tags} #{similarity || 'n/a'}>"
+    end
+
+    def tags_set
+      Ollama::Utils::Tags.new(tags, source:)
+    end
+
+    def ==(other)
+      text == other.text
+    end
+
+    alias inspect to_s
+  end
+
+  module RedisFullEach
+    def full_each(&block)
+      redis.scan_each(match: [ Ollama::Documents, ?* ] * ?-) do |key|
+        value = redis.get(key) or next
+        value = JSON(value, object_class: Ollama::Documents::Record)
+        block.(key, value)
+      end
+    end
+  end
+
+  module FindRecords
+    def find_records(needle, tags: nil, max_records: nil)
+      tags = Ollama::Utils::Tags.new(Array(tags)).to_a
+      records = self
+      if tags.present?
+        records = records.select { |_key, record| (tags & record.tags).size >= 1 }
+      end
+      needle_norm = norm(needle)
+      records     = records.sort_by { |key, record|
+        record.key        = key
+        record.similarity = cosine_similarity(
+          a: needle,
+          b: record.embedding,
+          a_norm: needle_norm,
+          b_norm: record.norm,
+        )
+      }
+      records.transpose.last&.reverse.to_a
+    end
+  end
+
+  module Tags
+    def clear(tags: nil)
+      tags = Ollama::Utils::Tags.new(tags).to_a
+      if tags.present?
+        if respond_to?(:clear_for_tags)
+          clear_for_tags(tags)
+        else
+          each do |key, record|
+            if (tags & record.tags.to_a).size >= 1
+              delete(unpre(key))
+            end
+          end
+        end
+      else
+        super()
+      end
+    end
+
+    def tags
+      if defined? super
+        super
+      else
+        each_with_object(Ollama::Utils::Tags.new) do |(_, record), t|
+          record.tags.each do |tag|
+            t.add(tag, source: record.source)
+          end
+        end
+      end
+    end
+  end
+end

data/lib/ollama/documents/cache/redis_backed_memory_cache.rb

@@ -5,8 +5,9 @@ class Ollama::Documents
     def initialize(prefix:, url: ENV['REDIS_URL'], object_class: nil)
       super(prefix:)
       url or raise ArgumentError, 'require redis url'
-      @prefix, @url, @object_class = prefix, url, object_class
+      @url, @object_class = url, object_class
       @redis_cache  = Ollama::Documents::RedisCache.new(prefix:, url:, object_class:)
+      @redis_cache.extend(Ollama::Documents::Cache::Records::RedisFullEach)
       @redis_cache.full_each do |key, value|
         @data[key] = value
       end

data/lib/ollama/documents/cache/redis_cache.rb

@@ -5,8 +5,9 @@ class Ollama::Documents::RedisCache
   include Ollama::Documents::Cache::Common
 
   def initialize(prefix:, url: ENV['REDIS_URL'], object_class: nil, ex: nil)
+    super(prefix:)
     url or raise ArgumentError, 'require redis url'
-    @prefix, @url, @object_class, @ex = prefix, url, object_class, ex
+    @url, @object_class, @ex = url, object_class, ex
   end
 
   attr_reader :object_class
@@ -18,7 +19,7 @@ class Ollama::Documents::RedisCache
   def [](key)
     value = redis.get(pre(key))
     unless value.nil?
-      JSON(value, object_class:)
+      object_class ? JSON(value, object_class:) : JSON(value)
     end
   end
 
@@ -64,12 +65,4 @@ class Ollama::Documents::RedisCache
     self
   end
   include Enumerable
-
-  def full_each(&block)
-    redis.scan_each(match: [ Ollama::Documents, ?* ] * ?-) do |key|
-      value = redis.get(key) or next
-      value = JSON(value, object_class: Ollama::Documents::Record)
-      block.(key, value)
-    end
-  end
 end