IBM Granite Knowledge

Signed-off-by: I Barnard <[email protected]>

Author: barny

knowledge/technology/attribution.txt

@@ -0,0 +1,5 @@
+Title of work: IBM Granite
+Link to work: https://en.wikipedia.org/wiki/IBM_Granite
+Revision: https://en.wikipedia.org/w/index.php?title=IBM_Granite&oldid=1269265317
+License of the work: CC-BY-SA-4.0
+Creator names: Wikipedia Staff

knowledge/technology/qna.yaml

@@ -0,0 +1,84 @@
+created_by: barny
+version: 3
+domain: large-language-model
+document_outline: Knowledge contribution about the IBM Granite Model
+seed_examples:
+  - context: >-
+      IBM Granite is a series of decoder-only AI foundation models created by
+      IBM.[3] It was announced on September 7, 2023,[4][5] and an initial paper
+      was published 4 days later.[6] Initially intended for use in the IBM's
+      cloud-based data and generative AI platform Watsonx along with other
+      models,[7] IBM opened the source code of some code models.[8][9] Granite
+      models are trained on datasets curated from Internet, academic
+      publishings, code datasets, legal and finance documents
+    questions_and_answers:
+      - question: What is IBM Granite?
+        answer: >-
+          IBM Granite is a series of decoder-only AI foundation models created
+          by IBM.
+      - question: When as was IBM Granite announced?
+        answer: September 7, 2023
+      - question: What's a series of IBM decoder-only AI foundation models?
+        answer: IBM Granite
+  - context: >-
+      Granite's first foundation models were Granite.13b.instruct and
+      Granite.13b.chat. The "13b" in their name comes from 13 billion, the
+      amount of parameters they have as models, lesser than most of the larger
+      models of the time. Later models vary from 3 to 34 billion parameters.
+    questions_and_answers:
+      - question: what were the first foundation models in IBm Granite?
+        answer: Granite.13b.instruct and Granite.13b.chat
+      - question: >-
+          How many model parameters do Granite.13b.instruct and Granite.13b.chat
+          have?
+        answer: 13 billion
+      - question: Do all Granite models have the same number of model parameters?
+        answer: Models vary from 3 to 34 billion parameters.
+  - context: >
+      On May 6, 2024, IBM released the source code of four variations of Granite
+      Code Models under Apache 2, an open source permissive license that allows
+      completely free use, modification and sharing of the software, and put
+      them on Hugging Face for public use.[14][15] According to IBM's own
+      report, Granite 8b outperforms Llama 3 on several coding related tasks
+      within similar range of parameters.
+    questions_and_answers:
+      - question: When did IBM first release the source code of Granite Code Models?
+        answer: May 6, 2024
+      - question: Does Granite outperform other LLMs?
+        answer: >-
+          Granite 8b outperforms Llama 3 on several coding related tasks within
+          similar range of parameters
+      - question: Where are the Granite models and code accessed?
+        answer: IBM put them on Hugging Face for public use.
+  - context: >-
+      A foundation model is an AI model trained on broad data at scale such that
+      it can be adapted to a wide range of downstream tasks
+    questions_and_answers:
+      - question: what is a foundation model?
+        answer: 'A foundation model is an AI model trained on broad data '
+      - question: What scale of data is a foundation model trained on?
+        answer: >-
+          the data is at scale such that it can be adapted to a wide range of
+          downstream tasks
+      - question: why are these tasks referred to as downstream?
+        answer: >-
+          "Downstream" means "derived from" or "based on"; downstream tasks are
+          work that is done using the foundation model.
+  - context: >-
+      IBM opened the source code of some code models.[8][9] Granite models are
+      trained on datasets curated from Internet, academic publishings, code
+      datasets, legal and finance documents
+    questions_and_answers:
+      - question: Did IBM release the full source code for IBM Granite?
+        answer: IBM released source code of four code models.
+      - question: What datasets is Granite trained on?
+        answer: >-
+          As well as datasets curated from the internet,  Granite is trained on
+          academic publications, code datasets, legal and finance documents
+      - question: Are the Granite training datasets open source?
+        answer: 'Yes'
+document:
+  repo: https://github.com/barny/taxonomy-knowledge-docs
+  commit: 260ddda8c25526f200a5fe60e3898cf303e2430c
+  patterns:
+    - granite-20250123T172453388.md