agentic-github-tagger
Lightweight text2text tag generator for agentic AI / RAG / LLMOps GitHub-style
descriptions. Fine-tuned from google-t5/t5-small
with PEFT LoRA (r=16, alpha=32, dropout=0.05, target_modules q,v) on
hharsha/agentic-github-meta,
then merged so full small weights load on free CPU.
~60M-param T5-small tagger — not a 7B chat demo. Free Hub + CPU friendly.
Usage
from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
model_id = "hharsha/agentic-github-tagger"
tok = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForSeq2SeqLM.from_pretrained(model_id)
text = "multi-agent platform with RAG, MCP, and observability"
ids = tok(text, return_tensors="pt")
out = model.generate(**ids, max_new_tokens=64, num_beams=4)
print(tok.decode(out[0], skip_special_tokens=True))
On older transformers that still register the task, this also works:
from transformers import pipeline
pipe = pipeline("text2text-generation", model="hharsha/agentic-github-tagger")
print(pipe("multi-agent platform with RAG, MCP, and observability")[0]["generated_text"])
Sample output from this training run:
multi-agent, multi-agent, observability, rag, mCP, observability
Training
| Base | google-t5/t5-small |
| Method | PEFT LoRA then merge |
| r / alpha / dropout | 16 / 32 / 0.05 |
| target_modules | q, v |
| Epochs | 3 (CPU) |
| Batch size | 8 |
| Dataset | hharsha/agentic-github-meta (687 rows; 600 used for train) |
Links
- Dataset:
hharsha/agentic-github-meta - Showcase:
hharsha/agentic-systems-showcase - Studio: https://agentic-systems-studio.com
- GitHub: https://github.com/hharsha98
Intended use / limits
Auto-suggest comma-separated tags for agentic / RAG / LLMOps project listings. Small model; tags can repeat or be incomplete. Not for safety-critical labeling.