llama_cpp 0.2.1 → 0.2.2

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: ad6a2964cfc46e940026d76a3d340509ba8c30fdaf3902730081f44b7b40cfde
-  data.tar.gz: 48384234163db26b7ee45d12310ba09b1a8f4f37906ede2f9d89eb72f05df665
+  metadata.gz: e5e221d4831be790a990b121e6ac780d10b4cbfb85b2a9b4284d9c216f6e5604
+  data.tar.gz: fba76ac1a70bfd7b02b8d123c57e4c8096a29ac7f658bb090cda91c6a54752d2
 SHA512:
-  metadata.gz: 132095fecc385ca629dc051d27bafddccf893def0702690abcaf7c3b87900c643ff301bf5f3f27db99a5c58ecb90385210e35c935cf2bd99f00b2675374b31c8
-  data.tar.gz: 5987962a6d84cdf7e7a171be41e7df96a0dab94d54f408df20303d4d1622ea851c6367d9773d4d985eaa1ba77f804ab730580a1a0a4374e96b5153c1a2471ed1
+  metadata.gz: 994029383219077e134d170177954251c20ede6d1c83843ecd22c42eeae83584079d124b41702f55add7f3f237e9bdb14382fbd37dde2d0e74f8cffcfed1715b
+  data.tar.gz: ca4e94b6ddf4e4e9ddabbb2b8309cf4b2b06a881df09fdf4ad96e27c4f1f620ca0024ac46f69d9b474849c074a5c9ba9b0440777a0b52a12413bc356457a02f3

data/CHANGELOG.md

@@ -1,3 +1,7 @@
+## [[0.2.2](https://github.com/yoshoku/llama_cpp.rb/compare/v0.2.1...v0.2.2)] - 2023-06-24
+
+- Bump bundled llama.cpp from master-a09f919 to master-7487137.
+
 ## [[0.2.1](https://github.com/yoshoku/llama_cpp.rb/compare/v0.2.0...v0.2.1)] - 2023-06-17
 
 - Bump bundled llama.cpp from master-4de0334 to master-a09f919.

data/examples/README.md

@@ -58,3 +58,35 @@ Options:
 
 Start chat
 ```
+
+## embedding.rb
+
+### Usage
+
+```sh
+$ cd examples
+$ gem install llama_cpp thor
+$ ./embedding.rb -m /path/to/quantized-model.bin -t 4 -p 'Hello, World.'
+...
+0.7191136479377747 0.5564611554145813 1.4210394620895386 -1.4874695539474487
+```
+
+### Options
+
+```
+$ ./embedding.rb help main
+Usage:
+  embedding.rb main -m, --model=MODEL -p, --prompt=PROMPT
+
+Options:
+  -s, [--seed=N]          # random seed
+                          # Default: -1
+  -t, [--threads=N]       # number of threads
+                          # Default: 2
+  -m, --model=MODEL       # path to model file
+  -p, --prompt=PROMPT     # prompt to generate embedding
+      [--n-gpu-layers=N]  # number of layers on GPU
+                          # Default: 0
+
+Extract embedding from prompt
+```

data/examples/embedding.rb

@@ -0,0 +1,37 @@
+#!/usr/bin/env ruby
+# frozen_string_literal: true
+
+# embedding.rb extracts embedding from prompt.
+# It is created with reference to embedding.cpp in llama.cpp examples:
+# - https://github.com/ggerganov/llama.cpp/blob/master/examples/embedding/embedding.cpp
+
+require 'llama_cpp'
+require 'thor'
+
+class Embedding < Thor # rubocop:disable Style/Documentation
+  default_command :main
+  desc 'main', 'Extract embedding from prompt'
+  option :seed, type: :numeric, aliases: '-s', desc: 'random seed', default: -1
+  option :threads, type: :numeric, aliases: '-t', desc: 'number of threads', default: 2
+  option :model, type: :string, aliases: '-m', desc: 'path to model file', required: true
+  option :prompt, type: :string, aliases: '-p', desc: 'prompt to generate embedding', required: true
+  option :n_gpu_layers, type: :numeric, desc: 'number of layers on GPU', default: 0
+  def main # rubocop:disable Metrics/AbcSize
+    params = LLaMACpp::ContextParams.new
+    params.seed = options[:seed]
+    params.n_gpu_layers = options[:n_gpu_layers]
+    params.embedding = true
+    context = LLaMACpp::Context.new(model_path: options[:model], params: params)
+
+    embd_input = context.tokenize(text: options[:prompt], add_bos: true)
+
+    return unless embd_input.size.positive?
+
+    context.eval(tokens: embd_input, n_past: 0, n_threads: options[:threads])
+
+    context.embeddings.each { |val| print("#{val} ") }
+    print("\n")
+  end
+end
+
+Embedding.start(ARGV)