metadata
license: apache-2.0
base_model: Jackrong/Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled
tags:
- reasoning
- code-review
- deployment
- fine-tuned
- lora
datasets:
- nohurry/Opus-4.6-Reasoning-3000x-filtered
pipeline_tag: text-generation
Alkaid A
Fine-tuned from Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled with a custom multi-phase code review, debugging, and deployment workflow.
Capabilities
- Structured code review with pros/cons analysis
- 5-iteration debug cycle with variations
- Production deployment strategy generation
- Security, scalability, and compliance deep dives
- Automated versioning (00.00.XX) and GitHub release management
- Documentation generation and test automation
Alkaid A Workflow
- Detailed code/plan feedback with pros and cons
- Guided debug phase
- Production deployment strategy
- 5x debug iterations with variations
- Security, scalability, compliance deep dive
- API endpoint testing and monitoring
- Help doc scraping and compatibility checks
- GitHub versioned releases (00.00.XX)
- Guided repository push
- User testing, benchmarking, hardening
- Developer documentation and automated tests
- Progress summary and acknowledgment
Training
- Method: LoRA SFT (rank 16, alpha 16)
- Data: 2,326 Opus reasoning traces + custom workflow examples
- Quantization: 4-bit NF4 during training
- Framework: Transformers + PEFT + TRL
- Base Model: Jackrong/Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled
Usage
from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained("pennydoesdev/Alkaid-A")
tokenizer = AutoTokenizer.from_pretrained("pennydoesdev/Alkaid-A")
Training Studio
Train your own version at: Alkaid-A-Studio
License
Apache 2.0