Mistral-7B-CauseAnalyzer

Model Card for Model ID

<!-- Provide a quick summary of what the model is/does. -->

The Mistral 7B - Cause Analyzer is a fine-tuned large language model designed for analyzing server logs, categorizing errors, and providing debugging solutions. It is optimized for predictive maintenance tasks and can be integrated into tools like Splunk or Grafana for real-time operational insights.

Model Details

Model Description

<!-- Provide a longer summary of what this model is. -->

This model was fine-tuned on real-world and synthetic log data from Esperanto servers using the LoRA technique. It excels in automating error categorization and debugging recommendations, reducing manual intervention and improving server health monitoring.

• Developed by: [Sivakrishna Yaganti, Shankar Jayaratnam]
• Funded by [optional]: [Esperanto Technologies]
• Shared by [optional]: [Sivakrishna Yaganti]
• Model type: [Casual language model]
• Finetuned from model [optional]: [Mistral 7B]

Model Sources [optional]

<!-- Provide the basic links for the model. -->

• Repository: https://huggingface.co/Esperanto/Mistral-7B-CauseAnalyzer

Uses

<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->

Direct Use

<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->

The model can be used to analyze server logs for error categorization and debugging without additional fine-tuning. It is suitable for:

1. Identifying patterns in server logs.
2. Automating the process of error categorization.
3. Generating debugging recommendations.

Downstream Use [optional]

<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->

[More Information Needed]

Out-of-Scope Use

<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->

1. The model is not intended for general text generation tasks unrelated to server log analysis.
2. It may not perform well on logs from domains significantly different from the training data.

Bias, Risks, and Limitations

<!-- This section is meant to convey both technical and sociotechnical limitations. -->

Bias:

1. The model's performance is optimized for logs similar to those in the training data. Logs with substantially different formats or languages may yield suboptimal results.

Risks:

1. Over-reliance on model predictions without validation could lead to incorrect debugging actions.
2. The model may fail to identify new or rare errors that were not part of the training data.

Limitations:

1. The model assumes logs are in English.
2. It may struggle with incomplete or highly noisy log data.

Recommendations

<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->

1. Validate predictions with domain experts, especially in critical systems.
2. Use the model alongside traditional debugging methods to ensure accuracy.

How to Get Started with the Model

Use the code below to get started with the model.

Load the model and tokenizer

• model_name = "Esperanto/Mistral-7B-CauseAnalyzer"
• tokenizer = AutoTokenizer.from_pretrained(model_name)
• model = AutoModelForCausalLM.from_pretrained(model_name)

Training Details

Training Data

<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->

• Source: Real-world logs from Esperanto servers augmented with synthetic logs generated using GPT-4.
• Size: ~170 labeled samples after data augmentation.

Training Procedure

<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->

Preprocessing [optional]

1. Logs were structured into fields for error type, root cause, and debugging solution.
2. Missing labels were generated using GPT-4 and manual verification.

• Fine-tuning method: LoRA (Low-Rank Adaptation)

Training Hyperparameters

• Training regime: [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->

Speeds, Sizes, Times [optional]

<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->

[More Information Needed]

Evaluation

<!-- This section describes the evaluation protocols and provides the results. -->

Testing Data, Factors & Metrics

Testing Data

<!-- This should link to a Dataset Card if possible. -->

Validation set: 10% of labeled data.

Factors

<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->

Model performance was evaluated on:

1. Error categorization accuracy.
2. Cause similarity score (cosine similarity) between predicted and ground truth causes.

Metrics

<!-- These are the evaluation metrics being used, ideally with a description of why. -->

Cause (Similarity Score):

1. Baseline Mistral 7B: 51.91
2. Mistral-7B-CauseAnalyzer: 67.15

Error Categorization Accuracy:

1. Baseline Mistral 7B: 46.23%
2. Mistral-7B-CauseAnalyzer: 70%

Results

Training and Validation Loss

1. Training Loss decreased steadily from ~1 to 0.38, as shown in the train/loss graph.
2. Evaluation Loss reduced from 0.6 to 0.3, indicating effective generalization.

Summary

The Fine-Tuned Mistral 7B - Cause Analyzer significantly outperforms the baseline models, achieving:

1. A 67.15 similarity score for cause prediction.
2. A 70% accuracy in error categorization.

These results highlight the model's robustness in predictive maintenance tasks and its potential for real-world integration into server health monitoring systems. -Had it been finetuned with more data, could have given better results.

Model Examination [optional]

<!-- Relevant interpretability work for the model goes here -->

[More Information Needed]

Environmental Impact

<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->

Carbon emissions can be estimated using the Machine Learning Impact calculator presented in Lacoste et al. (2019).

• Hardware Type: [More Information Needed]
• Hours used: [More Information Needed]
• Cloud Provider: [More Information Needed]
• Compute Region: [More Information Needed]
• Carbon Emitted: [More Information Needed]

Technical Specifications [optional]

Model Architecture and Objective

• Architecture: Mistral 7B causal language model.
• Objective: Fine-tuned for error categorization and debugging solutions in server logs.

Compute Infrastructure

[More Information Needed]

Hardware

Nvidia A100 and Esperanto Accelerators

Software

Hugging Face Transformers library.

Citation [optional]

<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->

BibTeX:

[More Information Needed]

APA:

[More Information Needed]

Glossary [optional]

<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->

[More Information Needed]

More Information [optional]

[More Information Needed]

Model Card Authors [optional]

Sivakrishna Yaganti, Shankar Jayaratnam

Model Card Contact

[email protected]