Update README.md

README.md

@@ -15,6 +15,7 @@ tags:
   - instruct
   - lightweight
   - safetensors
+  - withinusai
 
 license: other
 license_name: withinusai-custom-license
@@ -28,14 +29,158 @@ datasets:
   - TeichAI/gpt-5.1-codex-max-1000x
   - TeichAI/gpt-5.1-high-reasoning-1000x
 
-# Metrics are the names of metrics you will report (no fake values)
+# Metrics you intend to report (no fake values)
 metrics:
   - pass@1
+  - accuracy
   - exact_match
+
+# Eval Results container (empty until real numbers are added)
+model-index:
+  - name: WithinUsAI/GPT2.5.2-high-reasoning-codex-0.4B
+    results: []
+---
+---
+language:
+  - en
+
+library_name: transformers
+pipeline_tag: text-generation
+
+tags:
+  - gpt2
+  - causal-lm
+  - text-generation
+  - code
+  - coding
+  - reasoning
+  - instruct
+  - lightweight
+  - safetensors
+  - withinusai
+
+license: other
+license_name: withinusai-custom-license
+license_link: LICENSE
+
+base_model: openai-community/gpt2-medium
+base_model_relation: finetune
+
+datasets:
+  - WithinUsAI/GPT-2-to-GPT-5-5k
+  - TeichAI/gpt-5.1-codex-max-1000x
+  - TeichAI/gpt-5.1-high-reasoning-1000x
+
+# Metrics you intend to report (no fake values)
+metrics:
+  - pass@1
   - accuracy
+  - exact_match
 
-# Eval results metadata (kept empty until you add real numbers)
+# Eval Results container (empty until real numbers are added)
 model-index:
   - name: WithinUsAI/GPT2.5.2-high-reasoning-codex-0.4B
     results: []
 ---
+Prompting tips (high-reasoning unlock)
+
+“Give a 3-step plan, then implement.”
+
+“List edge cases first, then write the code.”
+
+“Explain root cause → propose fix → provide patch.”
+
+“State invariants + time/space complexity.”
+
+Fine-tuning and method (WithIn Us AI)
+
+WithIn Us AI created the concept, idea, and process behind enhancing GPT-2 Medium toward a GPT-5.2-style “twin target,” including:
+
+instruction/format design for reasoning + coding
+
+fine-tuning strategy and iteration process
+
+naming/versioning of the enhanced model line (“GPT2.5.2”)
+
+Base model credit
+
+This work is built on the foundation model:
+
+openai-community/gpt2-medium
+
+Training data (datasets used)
+
+This model is trained using a mixture of WithIn Us AI datasets plus third-party datasets used to improve training results:
+
+WithIn Us AI dataset (creator-owned):
+
+WithinUsAI/GPT-2-to-GPT-5-5k
+
+Third-party datasets (no ownership claimed, credited explicitly):
+
+TeichAI/gpt-5.1-codex-max-1000x
+
+TeichAI/gpt-5.1-high-reasoning-1000x
+
+Evaluation (add results when available)
+
+The metadata includes a model-index block so evaluation results can be added later and rendered by the Hub.
+
+Suggested benchmarks:
+
+Category	Benchmark	Metric
+Code	HumanEval	pass@1
+Code	MBPP	pass@1
+Reasoning	Custom reasoning set	accuracy
+Reliability	Bugfix set	fix rate
+Limitations
+
+May hallucinate library/API details; validate via execution
+
+Small models can struggle with long multi-hop constraints
+
+Always run unit tests and check edge cases
+
+Ethics & responsible use
+
+Outputs may reflect biases in training data
+
+Avoid using for harmful instructions or sensitive personal data
+
+Human review recommended for real deployments
+
+Thanks & attribution (explicit thanks by username + dataset/model name)
+
+WithIn Us AI proudly gives thanks and respect to the original creators whose work makes this ecosystem possible:
+
+Base model foundation (no ownership claimed):
+
+Thank you to the creators and maintainers of openai-community/gpt2-medium.
+
+Third-party datasets used (no remixing, no ownership claimed):
+
+Thank you TeichAI for TeichAI/gpt-5.1-codex-max-1000x.
+
+Thank you TeichAI for TeichAI/gpt-5.1-high-reasoning-1000x.
+
+WithIn Us AI original dataset:
+
+WithinUsAI/GPT-2-to-GPT-5-5k (created by WithIn Us AI)
+
+WithIn Us AI is excited about collaboration, research, experiments, and breakthrough discovery.
+
+License
+
+Licensed under the WithIn Us AI Custom License.
+See LICENSE.
+
+Citation
+@misc{withinusai_gpt252_high_reasoning_codex_04b,
+  title  = {WithinUsAI/GPT2.5.2-high-reasoning-codex-0.4B},
+  author = {WithIn Us AI},
+  year   = {2026},
+  url    = {https://huggingface.co/WithinUsAI/GPT2.5.2-high-reasoning-codex-0.4B}
+}
+Changelog
+
+v2.5.2: GPT-2 Medium enhanced toward GPT-5.2 twin target (reasoning + codex tuning)