Update model card

README.md

@@ -1,9 +1,71 @@
 ---
 license: apache-2.0
+language:
+- en
+pipeline_tag: text-generation
+datasets:
+- HuggingFaceTB/smollm-corpus
+- HuggingFaceTB/dclm-edu
+- HuggingFaceFW/finewiki
+- HuggingFaceFW/finephrase
+- allenai/dolmino-mix-1124
+- common-pile/stackv2_edu_filtered
+- HuggingFaceTB/smol-smoltalk
 tags:
 - petitgpt
 - native-pytorch
+- from-scratch
+- small-language-model
+- pretraining
+- sft
+- model-interpolation
+- gqa
+- single-gpu
 - research
+model-index:
+- name: petitgpt-alpha075
+  results:
+  - task:
+      type: text-generation
+      name: Text Generation
+    dataset:
+      type: allenai/ai2_arc
+      name: ARC-Easy
+      config: ARC-Easy
+      split: test
+      revision: 210d026faf9955653af8916fad021475a3f00453
+    metrics:
+    - type: acc
+      value: 57.74
+      name: accuracy (zero-shot likelihood, FP32)
+      verified: false
+    - type: acc_norm
+      value: 52.36
+      name: normalized accuracy (zero-shot likelihood, FP32)
+      verified: false
+    source:
+      name: PetitGPT technical report (native protocol-compatible evaluator; not an installed lm-evaluation-harness run)
+      url: https://github.com/yangqi0/petitgpt/blob/157ef969de38167dbf0dce04e9095344a4a4b177/docs/petitgpt-v1/TECHNICAL_REPORT.md
+  - task:
+      type: text-generation
+      name: Text Generation
+    dataset:
+      type: baber/piqa
+      name: PIQA
+      split: validation
+      revision: 142f6d7367fd9877f0fb3b5734ea6a545f54cdd1
+    metrics:
+    - type: acc
+      value: 63.49
+      name: accuracy (zero-shot likelihood, FP32)
+      verified: false
+    - type: acc_norm
+      value: 62.30
+      name: normalized accuracy (zero-shot likelihood, FP32)
+      verified: false
+    source:
+      name: PetitGPT technical report (native protocol-compatible evaluator; not an installed lm-evaluation-harness run)
+      url: https://github.com/yangqi0/petitgpt/blob/157ef969de38167dbf0dce04e9095344a4a4b177/docs/petitgpt-v1/TECHNICAL_REPORT.md
 ---
 
 # PetitGPT
@@ -115,3 +177,16 @@ The native export preserves all 213 named state entries and 60 non-persistent ro
 Author-controlled code, the model weights, and the tokenizer are released under [Apache-2.0](LICENSE). Author-written documentation is licensed separately under [CC BY 4.0](DOCUMENTATION_LICENSE.md). Third-party content retains its own terms and notices.
 
 The [source notice](SOURCE_NOTICE.md) records the dataset and collection identities, tokenizer linkage, differences between component notices, and publisher observations that reflect only current pages. See the [third-party notices](THIRD_PARTY_NOTICES.md), the [run guide](RUN_GUIDE.md), and the [commit-pinned technical report](https://github.com/yangqi0/petitgpt/blob/157ef969de38167dbf0dce04e9095344a4a4b177/docs/petitgpt-v1/TECHNICAL_REPORT.md) for the detailed research record.
+
+## Citation
+
+If you use this model or its documentation, please cite:
+
+```bibtex
+@misc{qi2026petitgpt,
+  title  = {PetitGPT: A 124.6M-Parameter Language Model Trained from Scratch on a Single GPU},
+  author = {Qi, Yang},
+  year   = {2026},
+  url    = {https://huggingface.co/yqi0/petitgpt}
+}
+```