From 643c0c5791a2fb9a4734798c295dcaadfd0cdff6 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bohuslav=20=C5=A0imek?= <bohuslav@simek.si>
Date: Sat, 22 Apr 2023 18:18:03 +0200
Subject: [PATCH] feat(LLM.php): add abstract call method to LLM class
 feat(llm-LLaMACpp.php): add example usage of LLaMACpp class with a prompt
 text

feat(LLaMACpp.php): add LLaMACpp class implementation with constructor, defaultParams, call, llmType, toArray, and getIdentifyingParams methods.
---
 README.md                 |   2 +-
 examples/llm-LLaMACpp.php |   9 ++
 src/LLMs/LLM.php          |  41 +++++++++
 src/LLMs/LLaMACpp.php     | 186 ++++++++++++++++++++++++++++++++++++++
 4 files changed, 237 insertions(+), 1 deletion(-)
 create mode 100644 examples/llm-LLaMACpp.php
 create mode 100644 src/LLMs/LLM.php
 create mode 100644 src/LLMs/LLaMACpp.php

diff --git a/README.md b/README.md
index e11f920..70aabe2 100644
--- a/README.md
+++ b/README.md
@@ -35,7 +35,7 @@ This library is still in development. Use at your own risk!
 ### Supported features
 
 * Prompt formatting
-* LLMs: OpenAI GPT-3, ChatGPT
+* LLMs: OpenAI GPT-3, ChatGPT, llama.cpp
 * Vector stores: Simple stupid vector store
 * Texts splitters: CharacterTextSplitter, RecursiveCharacterTextSplitter
 * Embeddings: OpenAI
diff --git a/examples/llm-LLaMACpp.php b/examples/llm-LLaMACpp.php
new file mode 100644
index 0000000..67811d3
--- /dev/null
+++ b/examples/llm-LLaMACpp.php
@@ -0,0 +1,9 @@
+<?php
+
+require __DIR__.'/../vendor/autoload.php';
+
+use Kambo\Langchain\LLMs\LLaMACpp;
+
+$llm = new LLaMACpp(['model_path' => 'path/to/model.bin']);
+$text = "What's the best programing language for web?";
+echo $llm($text);
diff --git a/src/LLMs/LLM.php b/src/LLMs/LLM.php
new file mode 100644
index 0000000..81b4b7c
--- /dev/null
+++ b/src/LLMs/LLM.php
@@ -0,0 +1,41 @@
+<?php
+
+namespace Kambo\Langchain\LLMs;
+
+/**
+ * LLM class that expect subclasses to implement a simpler call method.
+ * The purpose of this class is to expose a simpler interface for working
+ * with LLMs, rather than expect the user to implement the full generateResult method.
+ */
+abstract class LLM extends BaseLLM
+{
+    /**
+     * @param string $prompt
+     * @param array  $parameters
+     *
+     * @return string
+     */
+    abstract public function call(string $prompt, array $parameters = []): string;
+
+    /**
+     * Run the LLM on the given prompts.
+     * This is the method that should be implemented by subclasses ;-)
+     *
+     * @param array      $prompts
+     * @param array|null $stop
+     *
+     * @return LLMResult
+     */
+    public function generateResult(array $prompts, array $stop = null): LLMResult
+    {
+        $generations = [];
+        foreach ($prompts as $prompt) {
+            $result = $this->call($prompt, ['stop' => $stop]);
+            $generations[] = new Generation(
+                $result,
+            );
+        }
+
+        return new LLMResult([$generations], []);
+    }
+}
diff --git a/src/LLMs/LLaMACpp.php b/src/LLMs/LLaMACpp.php
new file mode 100644
index 0000000..0361678
--- /dev/null
+++ b/src/LLMs/LLaMACpp.php
@@ -0,0 +1,186 @@
+<?php
+
+namespace Kambo\Langchain\LLMs;
+
+use Kambo\Langchain\Callbacks\CallbackManager;
+use Psr\SimpleCache\CacheInterface;
+use Kambo\Langchain\Exceptions\InvalidArgumentException;
+use Kambo\LLamaCPPLangchainAdapter\LLamaCPPLangchainAdapter;
+
+use function class_exists;
+use function array_merge;
+
+/**
+ * Wrapper around the llama.cpp model.
+ */
+final class LLaMACpp extends LLM
+{
+    private $modelPath;
+    private $nCtx = 1024;
+    private $nParts = -1;
+    private $seed = 0;
+    private $f16Kv = false;
+    private $logitsAll = false;
+    private $vocabOnly = false;
+    private $useMlock = false;
+    private $nThreads = null;
+    private $nBatch = 8;
+    private $suffix = null;
+    private $maxTokens = 256;
+    private $temperature = 0.8;
+    private $topP = 0.95;
+    private $logprobs = null;
+    private $echo = false;
+    private $stop = [];
+    private $repeatPenalty = 1.1;
+    private $topK = 40;
+    private $lastNTokensSize = 64;
+    private LLamaCPPLangchainAdapter $adapter;
+
+   /**
+    * Possible options:
+    *
+    * model_path - The path to the Llama model file.
+    * n_ctx - Token context window.
+    * n_parts - Number of parts to split the model into. If -1, the number of parts is automatically determined.
+    * seed - Seed. If -1, a random seed is used.
+    * f16_kv - Use half-precision for key/value cache.
+    * logits_all - Return logits for all tokens, not just the last token.
+    * vocab_only - Only load the vocabulary, no weights.
+    * use_mlock - Force system to keep model in RAM.
+    * n_threads - Number of threads to use. If None, the number of threads is automatically determined.
+    * n_batch - Number of tokens to process in parallel. Should be a number between 1 and n_ctx.
+    * suffix - A suffix to append to the generated text. If None, no suffix is appended.
+    * max_tokens - The maximum number of tokens to generate.
+    * temperature - The temperature to use for sampling.
+    * top_p - The top-p value to use for sampling.
+    * logprobs - The number of logprobs to return. If None, no logprobs are returned.
+    * echo - Whether to echo the prompt.
+    * stop - A list of strings to stop generation when encountered.
+    * repeat_penalty - The penalty to apply to repeated tokens.
+    * top_k - The top-k value to use for sampling.
+    * last_n_tokens_size - The number of tokens to look back when applying the repeat_penalty.
+    *
+    * @param array                         $options
+    * @param CallbackManager|null          $callbackManager
+    * @param CacheInterface|null           $cache
+    * @param LLamaCPPLangchainAdapter|null $adapter
+    *
+    * @throws InvalidArgumentException
+    */
+    public function __construct(
+        array $options = [],
+        ?CallbackManager $callbackManager = null,
+        ?CacheInterface $cache = null,
+        ?LLamaCPPLangchainAdapter $adapter = null,
+    ) {
+        if (!class_exists(LLamaCPPLangchainAdapter::class)) {
+            throw new InvalidArgumentException(
+                'Could not found LLamaCPPLangchainAdapter.
+                Please install the LLamaCPPLangchainAdapter library to use this model.'
+            );
+        }
+
+        $this->modelPath = $options['model_path'] ?? $this->modelPath;
+        $this->nCtx = $options['n_ctx'] ?? $this->nCtx;
+        $this->nParts = $options['n_parts'] ?? $this->nParts;
+        $this->seed = $options['seed'] ?? $this->seed;
+        $this->f16Kv = $options['f16_kv'] ?? $this->f16Kv;
+        $this->logitsAll = $options['logits_all'] ?? $this->logitsAll;
+        $this->vocabOnly = $options['vocab_only'] ?? $this->vocabOnly;
+        $this->useMlock = $options['use_mlock'] ?? $this->useMlock;
+        $this->nThreads = $options['n_threads'] ?? $this->nThreads;
+        $this->nBatch = $options['n_batch'] ?? $this->nBatch;
+        $this->suffix = $options['suffix'] ?? $this->suffix;
+        $this->maxTokens = $options['max_tokens'] ?? $this->maxTokens;
+        $this->temperature = $options['temperature'] ?? $this->temperature;
+        $this->topP = $options['top_p'] ?? $this->topP;
+        $this->logprobs = $options['logprobs'] ?? $this->logprobs;
+        $this->echo = $options['echo'] ?? $this->echo;
+        $this->stop = $options['stop'] ?? $this->stop;
+        $this->repeatPenalty = $options['repeat_penalty'] ?? $this->repeatPenalty;
+        $this->topK = $options['top_k'] ?? $this->topK;
+        $this->lastNTokensSize = $options['last_n_tokens_size'] ?? $this->lastNTokensSize;
+
+        parent::__construct($options, $callbackManager, $cache);
+
+        if ($adapter === null) {
+            $adapter = LLamaCPPLangchainAdapter::create(
+                [
+                    'model_path' => $this->modelPath,
+                    'n_ctx' => $this->nCtx,
+                    'n_parts' => $this->nParts,
+                    'seed' => $this->seed,
+                    'f16_kv' => $this->f16Kv,
+                    'logits_all' => $this->logitsAll,
+                    'vocab_only' => $this->vocabOnly,
+                    'use_mlock' => $this->useMlock,
+                    'n_threads' => $this->nThreads,
+                ]
+            );
+        }
+
+        $this->adapter = $adapter;
+    }
+
+    /**
+     * Returns the default parameters for the model.
+     * @return array
+     */
+    private function defaultParams()
+    {
+        return [
+            'suffix' => $this->suffix,
+            'max_tokens' => $this->maxTokens,
+            'temperature' => $this->temperature,
+            'top_p' => $this->topP,
+            'logprobs' => $this->logprobs,
+            'echo' => $this->echo,
+            'stop_sequences' => $this->stop,
+            'repeat_penalty' => $this->repeatPenalty,
+            'top_k' => $this->topK,
+        ];
+    }
+
+    /**
+     * Call the Llama model and return the output.
+     *
+     * @param string $prompt The prompt to use for generation.
+     * @param mixed  $stop   A list of strings to stop generation when encountered.
+     *
+     * @return string The generated text.
+     */
+    public function call(string $prompt, $stop = null): string
+    {
+        $params = $this->defaultParams();
+
+        if ($this->stop && $stop !== null) {
+            throw new InvalidArgumentException('`stop` found in both the input and default params.');
+        } elseif ($this->stop) {
+            $params['stop_sequences'] = $this->stop;
+        } elseif ($stop !== null) {
+            $params['stop_sequences'] = $stop;
+        } else {
+            $params['stop_sequences'] = [];
+        }
+
+        $text = $this->adapter->predict($prompt, $params);
+
+        return $text;
+    }
+
+    public function llmType(): string
+    {
+        return 'llama.cpp';
+    }
+
+    public function toArray(): array
+    {
+        return $this->getIdentifyingParams();
+    }
+
+    public function getIdentifyingParams(): array
+    {
+        return array_merge(['model_path' => $this->modelPath], $this->defaultParams());
+    }
+}