Update Airoboros. Test look on preview site.

Author: DBlankvoort

airoboros.yaml

@@ -14,87 +14,87 @@ system:
     link: https://huggingface.co/jondurbin/airoboros-dpo-110b-3.3
     type: text
     performanceclass: full
-    basemodelname: Airoboros-110B-3.3
+    basemodelname: Qwen1.5-110B
     endmodelname: Airoboros-DPO-110B-3.3
     endmodellicense: Purposely left ambiguous
-    releasedate: 2023-07
-    notes:
+    releasedate: 2024-05
+    notes: Experimental model tuned primarily from synthetic data generated by the airoboros implementation of LLM self-instruction.
 
 org:
     name: Jon Durbin
     link: https://github.com/jondurbin
-    notes: Only active on GitHub since May 2023
+    notes: Primary contributor to chutes.ai.
 
 # availability:
 datasources_basemodel:
     class: closed
     link:
-    notes: Llama2 training data is nowhere documented or disclosed
+    notes: Base model data sources nowhere documented or specified.
 
 datasources_endmodel:
-    class: open
-    link: https://github.com/jondurbin/airoboros
-    notes: Airoboros, an implementation of the Self-Instruct paper
+    class: partial
+    link: https://huggingface.co/datasets/jondurbin/airoboros-3.2, https://huggingface.co/datasets/bluemoon-fandom-1-1-rp-cleaned, https://huggingface.co/datasets/boolq, https://huggingface.co/datasets/jondurbin/gutenberg-dpo-v0.1, https://huggingface.co/datasets/LDJnr/Capybara, https://huggingface.co/datasets/jondurbin/cinematika-v0.1, https://huggingface.co/datasets/glaiveai/glaive-function-calling-v2, https://huggingface.co/datasets/grimulkan/LimaRP-augmented, https://huggingface.co/datasets/piqa, https://huggingface.co/datasets/Vezora/Tested-22k-Python-Alpaca, https://huggingface.co/datasets/mattpscott/airoboros-summarization, https://huggingface.co/datasets/unalignment/toxic-dpo-v0.2
+    notes: Most data generated by airoboros, an implementation of the Self-Instruct paper. Many other data sources linked. One data source links to a 404 page. Additional data sources linked in tags but not outlined in model card.
 
 weights_basemodel:
-    class: partial
-    link:
-    notes: Llama2, made conditionally available by Meta
+    class: open
+    link: https://huggingface.co/Qwen/Qwen1.5-110B
+    notes: Weights available through HuggingFace.
 
 weights_endmodel:
     class: open
-    link: https://huggingface.co/jondurbin/airoboros-l2-70b-gpt4-1.4.1/tree/main
-    notes: Made available through HuggingFace
+    link: https://huggingface.co/jondurbin/airoboros-dpo-110b-3.3
+    notes: Weights available through HuggingFace.
 
 trainingcode:
     class: partial
-    link: https://gist.github.com/jondurbin/87fc040b92a3073125ed516b04bc6e19
-    notes: Repo exists for RL data but only a gist exists for model training and architecture
+    link: https://github.com/jondurbin/qlora
+    notes: Repo exists only for general fine-tuning implementation.
 
 # documentation:
 code:
     class: partial
-    link: https://github.com/jondurbin/airoboros/tree/main/airoboros
-    notes: What little code available is not very systematically documented
+    link: https://github.com/jondurbin/qlora
+    notes: Code is not very systematically documented.
 
 architecture:
-    class: partial
-    link: https://huggingface.co/jondurbin/airoboros-l2-70b-gpt4-1.4.1/discussions/2#64c29e4c617b36543dedac9a
-    notes: Some info can be gleaned at link but most remains undocumented
+    class: closed
+    link: 
+    notes: Earlier airoboros model versions included architecture information. Latest versions (>3.1.2) do not. 
 
 preprint:
     class: closed
     link:
-    notes: No preprint found
+    notes: No preprint found.
 
 paper:
     class: closed
     link:
-    notes: No peer-reviewed paper found
+    notes: No peer-reviewed paper found.
 
 modelcard:
     class: partial
-    link: https://huggingface.co/jondurbin/airoboros-65b-gpt4-1.4
-    notes: Instructs reader to look up model card for prior 65B Llama1 version
+    link: https://huggingface.co/jondurbin/airoboros-dpo-110b-3.3
+    notes: Model card contains some information, mainly relating to inference and licensing.
 
 datasheet:
-    class: partial
-    link: https://huggingface.co/datasets/jondurbin/airoboros-gpt4-1.4.1
-    notes: Datasheet for RL data only
+    class: closed
+    link: https://huggingface.co/datasets/jondurbin/airoboros-3.2, https://huggingface.co/datasets/bluemoon-fandom-1-1-rp-cleaned, https://huggingface.co/datasets/boolq, https://huggingface.co/datasets/jondurbin/gutenberg-dpo-v0.1, https://huggingface.co/datasets/LDJnr/Capybara, https://huggingface.co/datasets/jondurbin/cinematika-v0.1, https://huggingface.co/datasets/glaiveai/glaive-function-calling-v2, https://huggingface.co/datasets/grimulkan/LimaRP-augmented, https://huggingface.co/datasets/piqa, https://huggingface.co/datasets/Vezora/Tested-22k-Python-Alpaca, https://huggingface.co/datasets/mattpscott/airoboros-summarization, https://huggingface.co/datasets/unalignment/toxic-dpo-v0.2
+    notes: Datasheets contain differing levels of documentation. Exact data used as well as data collection and curation procedure unknown.
 
 # access:
 package:
-    class: open
+    class: partial
     link: https://pypi.org/project/airoboros/
-    notes: Pip package available.
+    notes: Pip package available, however only supports older versions of airoboros.
 
 api:
     class: closed
     link:
-    notes: No API found
+    notes: No API found.
     metaprompt: closed
 
 licenses:
     class: partial
-    link: https://huggingface.co/jondurbin/airoboros-l2-70b-gpt4-1.4.1#licence-and-usage-restrictions
-    notes: Licensing left ambiguous because of murky status of OpenAI-derived Self-Instruct data
+    link: https://huggingface.co/jondurbin/airoboros-dpo-110b-3.3#licence-and-usage-restrictions
+    notes: Licensing left ambiguous because of murky status of OpenAI-derived Self-Instruct data.