Introduction
Examples of Generative AI in Medicine are: Diagnostic assistance, Clinical documentation and Drug discovery; similar examples for finance using same technology are financial forecasting, investment analysis, risk evaluation etc. Advances in Large Language Models (LLMs) are enabling significantly improved decision-making automation across both domains by affording more accurate, manner- and context-aware answers that are processed with high efficiency.
However, many obstacles remain — from data privacy and model bias to the significant computational resources such as GPUs necessary for their operation. Solving this problem is the intent of models like Palmyra-Med and Palmyra-Fin, which provide domain specific expertise and better data management. The models are a leap in the use of AI effectively demonstrating how industry-specific innovations can be derived from specialized LLMs.
Who Developed These Models?
Writer, a firm famous for AI assistance services, proposed two modification models with prefixes Palmyra i.e. Palmyra-Med and Palmyra-Fin. The initiative was developed in collaboration with AI researchers and industry experts. In early 2020, Writer was established by AI experts with combined experience in language models and enterprise solutions. Their goal is to build an AI Toolkit for productivity and compliance across various industries. Writer specializes in crafting domain-specialist LLMs and has a laser focus on delivering ROI for firms.
Palmyra-Med and Palmyra-Fin
Commercially, the Palmyra-Med is an LLM tailored to a medical use case and has deployed into practice in multiple settings. The DSL models belong to Writer's domain-specific suite of LLMs. These are in the 70b variety, which is designed to output high power and not make you hate your life. These models are state-of-the-art in terms of medical and financial generative AI levels, even better than other pretrained counterparts such as GPT-4 and Med-PaLM-2.
Features of Palmyra-Med and Palmyra-Fin
Palmyra-Med:
- Extremely Accurate: 85.9% on medical benchmarks
- Strengths: High clinical knowledge, anatomical / genetic accuracy, biomedical research.
Palmyra-Fin:
- Evidence of Professional Ability: Qualified CFA Level III test-taking, 73%.
- In-Depth Analysis: Ability to analyse financial trends, assess investment valuations and risks.
Use Cases of Palmyra-Med and Palmyra-Fin
Palmyra-Med:
- Diagnostic Support: Analyzes patient's data and advises diagnoses with treatments.
- Medicine: Summarizes articles and their relevant data to increase the speed of discoveryemics.
- Clinical Knowledge and Anatomy: Offers a broad range of detailed descriptions for medical procedures, Exam-level Tips on Clinical Behavior which can be used in day-to-day clinical behavior.
- Genetics and Personalized Medicine: Utilize genetic information to recommend individual treatment plans.
Palmyra-Fin:
- Financial Forecasting: Explaining how the market is expected to move due to previous trends in market, along with economics fundamentals.
- Investment Analysis: Assesses companies and the market to recommend investments opportunities.
- Risk Rating: Articulates both risks related to financial instruments as well offers comprehensive risk analysis.
- Preferred Asset Allocation Strategy: Suggests diversified investment portfolios matched to individual risk tolerance.
- Fraud Detection: analyze transaction patterns and use advance algorithms to find irregular activities.
How Does Palmyra-Med and Palmyra-Fin Work?
Palmyra-Med and Palmyra-Fin—uses a full-stack, generative-AI platform to provide domain-specific, high-accuracy and high-compliance models for both medical and financial applications. The technologies at the core of these models are integrated graph-based RAG (Retrieval-Augmented Generation) technology and are going to help with processing and content generation on a domain-specific basis using state-of-the-art NLP techniques. This technology is combined with AI guardrails and a suite of developer tools so that the models are both accurate and also regulatory-compliant.
The graph-based RAG technology used in Palmyra-Med and Palmyra-Fin represents a major breakthrough over traditional vector-based RAG approaches. Writer said that this technology can enable scalable processing of data, and in building valuable semantic connections between the data points by using a specialized LLM, hence cost-effective and easily updatable in terms of graph structures. Thus, attuned models with relevance to data pertaining to any query are developed for minimum hallucinations and maximum performance. RAG Graph-based Technology: This is a version of Writer's Knowledge Graph RAG, unique with the best performance in terms of accuracy and in handling data at the enterprise level.
source - https://writer.com/blog/palmyra-med-fin-models/
Its workflow comprises processing the data, querying and retrieval, and answer generation. The entities and relationships, however, have to be represented in the form of nodes and edges in a graph-based RAG structure in a data processing step. The query and retrieval step are performed by a mixture of NLP algorithms, heuristic algorithms, and machine learning techniques applied to understand the context of a query and to identify pertinent entities and relationships. Finally, the answer generation step uses these retrieved data points to formulate an accurate and compliant response. With this advanced technology, Palmyra-Med and Palmyra-Fin provide real-time performance in medical and financial applications, which is perfect for developing informed and responsible AI apps.
Performance Evaluation
Performance of Palmyra-Med with many benchmarks, such as MMLU Clinical Knowledge, Professional Medicine, PubMedQA, and others. In general, for medical benchmarks, Palmyra-Med ended up with an 85.9% average, overperforming Med-PaLM2 by almost 2 percentage points. From the MMLU Clinical Knowledge benchmark, Palmyra-Med scores 90.9%, showing good and rich understanding in clinical procedures and the human anatomy. Such a unique achievement gives Palmyra-Med the same worthiness and trust for ensuring good diagnostic accuracy and effective planning of treatment within health care.
source - https://writer.com/blog/palmyra-med-fin-models/
Palmyra-Fin passed the test of the CFA Level III with long-fin-eval benchmark tests. However, more importantly, Palmyra-Fin scored a massive 73% on the multiple choice part of an actual CFA Level III mock exam and, thereby, became the first model to pass the examination. This is in itself an achievement, as passing the CFA Level III is among the most difficult distinctions in the investment management profession. To top this, Palmyra-Fin surpasses evaluation of tests conducted on popular models such as Claude 3.5 Sonnet, GPT4o, and Mixtral-8x7b in a long-fin-eval benchmark test, thereby completely demonstrating its competencies and financial know-how.
source - https://huggingface.co/Writer/Palmyra-Fin-70B-32K
In addition to these well-known evaluations, a number of tests and experiments have been set up to check the performance level of Palmyra-Med and Palmyra-Fin. These include appraisals on genetics and college medicine, biomedical research, and financial trend analysis and forecasts. The findings for such appraisals have been documented in the underlying tables and figures referenced by the document.
How to Access and Use These Models
It can be accessed in both Writer's API and No-code tools, as well as through the writer framework. They are accessible on platforms like by NVIDIA NIM or Hugging Face. These models can be deployed in users premises or private cloud with an open-model license. Also contact a Writer's team on their licensing if you want to use it for commercial usage. These models are purpose-built to be easily incorporated into your task-specific workflows for flexible, scalable use in a variety of applications. Users who wants to learn more about this AI models, all important links associated with this models are provided at end of this article.
Limitations and Future Work
Palmyra-Med and Palmyra-Fin suffer from a limitation on the guiding of performance by the training data; they are thus potentially vulnerable to biased or incorrect inputs. These requirements make these models very expensive and unscalable in practice. They cannot replace human judgment; in addition, human oversight would lower the risk of errors. The high accuracy and performance of these state-of-the-art models highly depend on the knowledge graph; hence, this entity is at risk by having to depend solely on accurate data.
These issues can be mitigated through developing data quality, availability, scalability, and efficiency optimizations, and integrating human judgment in decision-making. The Knowledge Graph will be continually developed with evaluation and refinement, including multimodal inputs for processing various data types. It will work on identification and minimization of sources of bias or unfairness in model predictions.
Conclusion
The Palmyra-Med and Palmyra-Fin versions rely on use case specific expertise to achieve the highest accuracy while handling today challenges like data privacy, model bias or high computational requirements. They embody how specialized LLMs can advance innovation and outcomes in highly regulated areas, providing critical implications and practical tools to professionals who work in the field. While this being the tip of the iceberg, it shows what AI can do in specific business sectors, how change is knocking down on a variety of doors.
Source
Blog: https://writer.com/blog/palmyra-med-fin-models/
Writer: https://writer.com/#palmyra
Knowledge base graph: https://writer.com/product/graph-based-rag/
Vector based retrieval limitation: https://writer.com/blog/vector-based-retrieval-limitations-rag/
Open model license: https://writer.com/legal/open-model-license/
HF model weights: https://huggingface.co/Writer
Disclaimer - This article is intended purely for informational purposes. It does not constitute legal, financial, medical, or professional advice. It is not sponsored or endorsed by any company or organization, nor does it serve as an advertisement or promotion for any product or service. All information presented is based on publicly available resources and is subject to change. Readers are encouraged to conduct their own research and due diligence.
No comments:
Post a Comment