issdandavis's picture
docs: add a card stating what is actually in this repo
1f9c2b6 verified
|
Raw
History Blame Contribute Delete
1.04 kB
---
license: apache-2.0
tags:
- polly
- scbe
- qwen2
- text-generation
- sft
- merged
- superseded
---
<!-- scbe-status -->
> **Status: superseded.** Research artifact, not a production candidate. Canonical model: [scbe-coding-agent-vtc-qwen15-v1-gguf](https://hf.co/issdandavis/scbe-coding-agent-vtc-qwen15-v1-gguf).
# Polly r8 - Qwen2.5-Coder-1.5B (merged)
Full-weight merge. Facts below are from `MERGE_INFO.txt` in this repo.
- **Base**: `Qwen/Qwen2.5-Coder-1.5B-Instruct`
- **Adapter**: `training/runs/local_1.5b_r8_20260415T202900/checkpoint-500`
- **Round**: r8 - 137K coding examples + tokenizer + spiral seal
- **Method**: `merge_and_unload` (LoRA baked into base weights)
## Important
The Kaggle 0.5B adapter at
[polly-r8-qwen-0.5b](https://hf.co/issdandavis/polly-r8-qwen-0.5b) is a
**different base model**. The two cannot be averaged or merged together.
## Not published here
No eval numbers are included in this repo, so none are claimed. Treat this as
a training artifact from round r8, not a benchmarked release.