archive: copy agent-q3-core README into canonical repo
Browse files
archive/agent-q3-core/README.md
ADDED
|
@@ -0,0 +1,25 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
# Agent Q3 [Evo] HFRepo
|
| 3 |
+
|
| 4 |
+
**Archived from:** `madDegen/agent-q3-core`
|
| 5 |
+
**Archived to canonical repo:** `madDegen/Agent-Q3`
|
| 6 |
+
|
| 7 |
+
## About
|
| 8 |
+
|
| 9 |
+
This README was originally the root README of the `madDegen/agent-q3-core` HuggingFace model repo,
|
| 10 |
+
labeled **Agent Q3 [Evo] HFRepo**.
|
| 11 |
+
|
| 12 |
+
Agent Q3 [Evo] is the self-improving evolution variant:
|
| 13 |
+
- Unsloth LoRA fine-tuning (SFTTrainer on Llama-3.2-3B-Instruct-bnb-4bit)
|
| 14 |
+
- arXiv ingestion pipeline → ChromaDB (384-dim nomic-embed-text)
|
| 15 |
+
- DPO/RLHF feedback collector
|
| 16 |
+
- LangGraph loop: ingest → train → benchmark → feedback (continues if score < 0.80)
|
| 17 |
+
- 4 domain benchmarks: prediction_markets / solidity / langgraph / lora
|
| 18 |
+
|
| 19 |
+
All code lives in the canonical repo at `madDegen/Agent-Q3` under the `evo/` directory.
|
| 20 |
+
|
| 21 |
+
## Links
|
| 22 |
+
- Canonical Code Repo: https://huggingface.co/madDegen/Agent-Q3
|
| 23 |
+
- Canonical Dataset/Bucket: https://huggingface.co/datasets/madDegen/agent-q3
|
| 24 |
+
- Canonical Space: https://huggingface.co/spaces/madDegen/agent-q3-space
|
| 25 |
+
- GitHub: https://github.com/MADdegen/Agent-Q3
|