v00.00.00: Initial model card for Alkaid A
Browse files
README.md
ADDED
|
@@ -0,0 +1,67 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
base_model: Jackrong/Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled
|
| 4 |
+
tags:
|
| 5 |
+
- reasoning
|
| 6 |
+
- code-review
|
| 7 |
+
- deployment
|
| 8 |
+
- fine-tuned
|
| 9 |
+
- lora
|
| 10 |
+
datasets:
|
| 11 |
+
- nohurry/Opus-4.6-Reasoning-3000x-filtered
|
| 12 |
+
pipeline_tag: text-generation
|
| 13 |
+
---
|
| 14 |
+
|
| 15 |
+
# Alkaid A
|
| 16 |
+
|
| 17 |
+
Fine-tuned from **Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled** with a custom
|
| 18 |
+
multi-phase code review, debugging, and deployment workflow.
|
| 19 |
+
|
| 20 |
+
## Capabilities
|
| 21 |
+
|
| 22 |
+
- Structured code review with pros/cons analysis
|
| 23 |
+
- 5-iteration debug cycle with variations
|
| 24 |
+
- Production deployment strategy generation
|
| 25 |
+
- Security, scalability, and compliance deep dives
|
| 26 |
+
- Automated versioning (00.00.XX) and GitHub release management
|
| 27 |
+
- Documentation generation and test automation
|
| 28 |
+
|
| 29 |
+
## Alkaid A Workflow
|
| 30 |
+
|
| 31 |
+
1. Detailed code/plan feedback with pros and cons
|
| 32 |
+
2. Guided debug phase
|
| 33 |
+
3. Production deployment strategy
|
| 34 |
+
4. 5x debug iterations with variations
|
| 35 |
+
5. Security, scalability, compliance deep dive
|
| 36 |
+
6. API endpoint testing and monitoring
|
| 37 |
+
7. Help doc scraping and compatibility checks
|
| 38 |
+
8. GitHub versioned releases (00.00.XX)
|
| 39 |
+
9. Guided repository push
|
| 40 |
+
10. User testing, benchmarking, hardening
|
| 41 |
+
11. Developer documentation and automated tests
|
| 42 |
+
12. Progress summary and acknowledgment
|
| 43 |
+
|
| 44 |
+
## Training
|
| 45 |
+
|
| 46 |
+
- **Method:** LoRA SFT (rank 16, alpha 16)
|
| 47 |
+
- **Data:** 2,326 Opus reasoning traces + custom workflow examples
|
| 48 |
+
- **Quantization:** 4-bit NF4 during training
|
| 49 |
+
- **Framework:** Transformers + PEFT + TRL
|
| 50 |
+
- **Base Model:** [Jackrong/Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled](https://huggingface.co/Jackrong/Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled)
|
| 51 |
+
|
| 52 |
+
## Usage
|
| 53 |
+
|
| 54 |
+
```python
|
| 55 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer
|
| 56 |
+
|
| 57 |
+
model = AutoModelForCausalLM.from_pretrained("pennydoesdev/Alkaid-A")
|
| 58 |
+
tokenizer = AutoTokenizer.from_pretrained("pennydoesdev/Alkaid-A")
|
| 59 |
+
```
|
| 60 |
+
|
| 61 |
+
## Training Studio
|
| 62 |
+
|
| 63 |
+
Train your own version at: [Alkaid-A-Studio](https://huggingface.co/spaces/pennydoesdev/Alkaid-A-Studio)
|
| 64 |
+
|
| 65 |
+
## License
|
| 66 |
+
|
| 67 |
+
Apache 2.0
|