add Glider to evals

tamingllms/_build/html/_sources/notebooks/evals.ipynb

@@ -802,6 +802,11 @@
     "While evaluating language model outputs inherently involves subjective judgment, establishing a high-quality benchmark model and using quantifiable metrics provide a more objective framework for comparing model performance. This approach transforms an otherwise qualitative assessment into a measurable, data-driven evaluation process.\n"
    ]
   },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": []
+  },
   {
    "cell_type": "markdown",
    "metadata": {},
@@ -810,6 +815,7 @@
     "\n",
     "*   **Task-specific nature**:  Chosen set of metrics might not fully capture the nuances of complex generative-based tasks, especially those involving subjective human judgment.\n",
     "*   **Sensitivity to data distribution**: Performance on these metrics can be influenced by the specific dataset used for evaluation, which might not represent real-world data distribution.\n",
+    "*   **Subjective Acceptable Threshold**: These metrics are not always easy to interpret and set a threshold for (see {cite}`sarmah2024choosethresholdevaluationmetric` for a discussion on how to choose a threshold for an evaluation metric for large language models).\n",
     "*   **Inability to assess reasoning or factual accuracy**: These metrics primarily focus on surface-level matching and might not reveal the underlying reasoning process of the LLM or its ability to generate factually correct information.\n",
     "\n",
     "In conclusion, selecting an appropriate extrinsic metrics set depends on the specific task, underlying business requirements and desired evaluation granularity.  Understanding the limitations of these metrics can provide a more comprehensive assessment of LLM performance in real-world applications.\n",
@@ -1143,12 +1149,11 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "Leveraging LLMs for evaluation has several limitations {cite}`li2024leveraginglargelanguagemodels`. Firstly, computational overhead should not be neglected given the inherent cost of running additional model inferences iterations. LLM evaluators can also exhibit various biases, including order bias (preferring certain sequence positions), egocentric bias (favoring outputs from similar models), and length bias. Further, there may be a tight dependency on prompt quality - small prompt variations may lead to substantially different outcomes. It is important to also note challenges around domain-specific evaluation in fields such as medice, finance, law etc, where a general llm-as-a-judge approach may not be suitable.\n",
-    "\n",
-    "The LLM-as-a-Judge strategy can serve as a scalable and nuanced solution to evaluate LLM-based applications. While it does not entirely a metrics-based or human-based aproach, it significantly augments evaluation workflows, especially in scenarios requiring evaluation of generative outputs. Future improvements could include integrating human oversight and refining LLMs for domain-specific evaluation tasks.\n",
-    "\n",
+    "Leveraging LLMs for evaluation has several limitations {cite}`li2024leveraginglargelanguagemodels`. Firstly, computational overhead should not be neglected given the inherent cost of running additional model inferences iterations. LLM evaluators can also exhibit various biases, including order bias (preferring certain sequence positions), egocentric bias (favoring outputs from similar models), and length bias. Further, there may be a tight dependency on prompt quality - small prompt variations may lead to substantially different outcomes. It is important to also note challenges around domain-specific evaluation in fields such as medicine, finance, law etc, where a general llm-as-a-judge approach may not be suitable.\n",
     "\n",
+    "The LLM-as-a-Judge strategy can serve as a scalable and nuanced solution to evaluate LLM-based applications. While it does not entirely replace metrics-based or human-based approaches, it significantly augments evaluation workflows, especially in scenarios requiring evaluation of generative outputs. Future improvements in our example include integrating human oversight and refining LLMs for domain-specific evaluation tasks.\n",
     "\n",
+    "One open source solution trying to overcome some of these challenges is Glider {cite}`deshpande2024glidergradingllminteractions`, a 3B evaluator LLM that can score any text input and associated context on arbitrary user defined criteria. Glider is an LLM model trained on 685 domains and 183 criteria whose judgement scores show 91.3% agreement with human judgments, making it suitable for a diverse range of real world applications.\n",
     "\n"
    ]
   },

tamingllms/_build/html/markdown/preface.html

@@ -236,15 +236,15 @@ <h1><span class="section-number">1. </span>Preface<a class="headerlink" href="#p
 <div><p>Models tell you merely what something is like, not what something is.</p>
 <p class="attribution">—Emanuel Derman</p>
 </div></blockquote>
-<p>An alternative title of this book could have been “Language Models Behaving Badly”. If you are coming from a background in financial modeling, you may have noticed the parallel with Emanuel Derman’s seminal work “Models.Behaving.Badly” <span id="id1">[<a class="reference internal" href="#id139" title="E. Derman. Models.Behaving.Badly.: Why Confusing Illusion with Reality Can Lead to Disaster, on Wall Street and in Life. Free Press, 2011. ISBN 9781439165010. URL: https://books.google.co.uk/books?id=lke_cwM4wm8C.">Derman, 2011</a>]</span>. This parallel is not coincidental. Just as Derman cautioned against treating financial models as perfect representations of reality, this book aims to highlight the limitations and pitfalls of Large Language Models (LLMs) in practical applications.</p>
+<p>An alternative title of this book could have been “Language Models Behaving Badly”. If you are coming from a background in financial modeling, you may have noticed the parallel with Emanuel Derman’s seminal work “Models.Behaving.Badly” <span id="id1">[<a class="reference internal" href="#id141" title="E. Derman. Models.Behaving.Badly.: Why Confusing Illusion with Reality Can Lead to Disaster, on Wall Street and in Life. Free Press, 2011. ISBN 9781439165010. URL: https://books.google.co.uk/books?id=lke_cwM4wm8C.">Derman, 2011</a>]</span>. This parallel is not coincidental. Just as Derman cautioned against treating financial models as perfect representations of reality, this book aims to highlight the limitations and pitfalls of Large Language Models (LLMs) in practical applications.</p>
 <p>The book “Models.Behaving.Badly” by Emanuel Derman, a former physicist and Goldman Sachs quant, explores how financial and scientific models can fail when we mistake them for reality rather than treating them as approximations full of assumptions.
 The core premise of his work is that while models can be useful tools for understanding aspects of the world, they inherently involve simplification and assumptions. Derman argues that many financial crises, including the 2008 crash, occurred partly because people put too much faith in mathematical models without recognizing their limitations.</p>
 <p>Like financial models that failed to capture the complexity of human behavior and market dynamics, LLMs have inherent constraints. They can hallucinate facts, struggle with logical reasoning, and fail to maintain consistency across long outputs. Their responses, while often convincing, are probabilistic approximations based on training data rather than true understanding even though humans insist on treating them as “machines that can reason”.</p>
 <p>Today, there is this growing pervasive belief that these models could solve any problem, understand any context, or generate any content as wished by the user. Moreover, language models that were initially designed to be next-token prediction machines and chatbots are now been twisted and wrapped into “reasoning” machines for further integration into technology products and daily-life workflows that control, affect, or decide daily actions of our lives. This technological optimism coupled with lack of understanding of the models’ limitations may pose risks we are still trying to figure out.</p>
 <p>This book serves as an introductory, practical guide for practitioners and technology product builders - software engineers, data scientists, and product managers - who want to create the next generation of GenAI-based products with LLMs while remaining clear-eyed about their limitations and therefore their implications to end-users. Through detailed technical analysis, reproducible Python code examples we explore the gap between LLM capabilities and reliable software product development.</p>
 <p>The goal is not to diminish the transformative potential of LLMs, but rather to promote a more nuanced understanding of their behavior. By acknowledging and working within their constraints, developers can create more reliable and trustworthy applications. After all, as Derman taught us, the first step to using a model effectively is understanding where it breaks down.</p>
 <div class="docutils container" id="id2">
-<div class="citation" id="id139" role="doc-biblioentry">
+<div class="citation" id="id141" role="doc-biblioentry">
 <span class="label"><span class="fn-bracket">[</span><a role="doc-backlink" href="#id1">Der11</a><span class="fn-bracket">]</span></span>
 <p>E. Derman. <em>Models.Behaving.Badly.: Why Confusing Illusion with Reality Can Lead to Disaster, on Wall Street and in Life</em>. Free Press, 2011. ISBN 9781439165010. URL: <a class="reference external" href="https://books.google.co.uk/books?id=lke_cwM4wm8C">https://books.google.co.uk/books?id=lke_cwM4wm8C</a>.</p>
 </div>