Add benchmark overview and training dataset associations

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+assets/overview.png filter=lfs diff=lfs merge=lfs -text

README.md

@@ -10,6 +10,20 @@ tags:
 - rlcd
 - lora
 - research
+datasets:
+- AmazonScience/massive
+- google/boolq
+- fancyzhx/ag_news
+- dair-ai/emotion
+- PolyAI/banking77
+- facebook/xnli
+- HuggingFaceM4/A-OKVQA
+- derek-thomas/ScienceQA
+- lmms-lab-encoder/VQAv2
+- LocalLLaMA/typed-decisions
+- microsoft/wiki_qa
+- UCLNLP/sharc
+- ucirvine/sms_spam
 ---
 
 # Dohnuts-0.1.0-0.8B
@@ -20,6 +34,8 @@ ordered scores. Independent questions about one input share computation in a
 single forward pass. The interface returns decisions without generating reasoning
 or free-form answers.
 
+![Dohnuts 0.1.0 model overview and benchmarks](https://huggingface.co/PsiACE/Dohnuts-0.1.0-0.8B/resolve/main/assets/overview.png)
+
 ## Model details
 
 | Property | Value |
@@ -130,6 +146,28 @@ samples perturb decision logits; they are not generated trajectories. LoRA is
 merged before temperature fitting and final evaluation. The
 [RLCD specification](https://github.com/PsiACE/dohnuts/blob/main/docs/rlcd.md) gives the objective and fixed schedule.
 
+### Training datasets
+
+The 26 training groups are derived from the following source datasets. Hub links
+identify the datasets; the [download manifests](https://github.com/PsiACE/dohnuts/tree/main/data/manifests/) pin the files,
+revisions, and checksums actually used, including official archives downloaded
+outside the Hub.
+
+| Task family | Sources |
+| --- | --- |
+| Intent and topic classification | [MASSIVE 1.1](https://huggingface.co/datasets/AmazonScience/massive) (en-US, zh-CN), [AG News](https://huggingface.co/datasets/fancyzhx/ag_news), [BANKING77](https://huggingface.co/datasets/PolyAI/banking77) |
+| Entailment, emotion, and Boolean QA | [XNLI](https://huggingface.co/datasets/facebook/xnli) (en, zh), [emotion](https://huggingface.co/datasets/dair-ai/emotion), [BoolQ](https://huggingface.co/datasets/google/boolq) (SuperGLUE distribution) |
+| Visual decisions | [CLEVR 1.0](https://cs.stanford.edu/people/jcjohns/clevr/), [A-OKVQA](https://huggingface.co/datasets/HuggingFaceM4/A-OKVQA), [ScienceQA](https://huggingface.co/datasets/derek-thomas/ScienceQA) (image subset), [VQAv2](https://huggingface.co/datasets/lmms-lab-encoder/VQAv2) (yes/no) |
+| Screen region decisions | [ScreenQA](https://github.com/google-research-datasets/screen_qa), with [Rico](https://www.interactionmining.org/archive/rico) screenshots and view hierarchies |
+| Typed decisions | [LocalLLaMA/typed-decisions](https://huggingface.co/datasets/LocalLLaMA/typed-decisions), using public soft teacher distributions |
+| Retrieval and relevance | [Amazon ESCI](https://github.com/amazon-science/esci-data) (en, es, ja), [WikiQA](https://huggingface.co/datasets/microsoft/wiki_qa) |
+| Policy and contract decisions | [ShARC](https://huggingface.co/datasets/UCLNLP/sharc), [ContractNLI](https://stanfordnlp.github.io/contract-nli/) |
+| Spam and phishing | [SpamAssassin](https://spamassassin.apache.org/old/publiccorpus/), [Nazario phishing corpus](https://monkey.org/~jose/phishing/), [UCI SMS Spam Collection](https://huggingface.co/datasets/ucirvine/sms_spam) |
+
+JevBench tasks and the frozen Laya benchmark inputs are evaluation-only.
+The [data protocol](https://github.com/PsiACE/dohnuts/blob/main/docs/data-and-evaluation.md) describes source-specific
+conversions, grouped partitions, exclusions, and terms.
+
 ## Limitations
 
 - Quality depends on the task. Jev leads on the public JevBench tasks; Laya

SHA256SUMS

@@ -1,5 +1,6 @@
 600ca4e25fe11762b75a97e714707fab48bb778374e92d24c6ca068791661c11  LICENSE
 f3b4840eb0568e827f4ba3096850aecbf5347a0fa3307542cf2b8e40ba9367d5  NOTICE
-6dec0a5d8401c41b0b0c8ba6146e8a306c2327d6fb2e67a8cc32486b5a03f6c9  README.md
+b2a8fb161d262d35bef46a3a2788ae50da77df8b19c15a2a4d5814866c34221c  README.md
 196be33a0282537bcd821e2115643b352d0ad2a0bbaf7b242a1b7fe5bd96cfdf  adapter.safetensors
 379b05ae34a1f5758a5ccec793b5b059273ab4f4496434f6edefeedd22e2baa4  dohnuts.json
+8b9cdf495fa2b5b70f05c3b015f4f1013e8cd5891be0b9cf8181590608558312  assets/overview.png