61 lines
2.0 KiB
Markdown
61 lines
2.0 KiB
Markdown
|
---
|
||
|
|
library_name: transformers
|
||
|
|
pipeline_tag: text-generation
|
||
|
|
base_model: Qwen/Qwen2.5-1.5B-Instruct
|
||
|
|
license: apache-2.0
|
||
|
|
tags:
|
||
|
|
- transformers
|
||
|
|
- safetensors
|
||
|
|
- text-generation
|
||
|
|
- qwen2
|
||
|
|
- lora
|
||
|
|
- merged-adapter
|
||
|
|
- brainalign
|
||
|
|
---
|
||
|
|
|
||
|
|
# C
|
||
|
|
|
||
|
|
## Summary
|
||
|
|
|
||
|
|
This model is the `C` BrainAlign stage-2 LoRA checkpoint (`best_by_retrieval`) merged into the full base model `Qwen/Qwen2.5-1.5B-Instruct`.
|
||
|
|
|
||
|
|
The folder is saved in standard Hugging Face format so it can be uploaded directly for Open LLM Leaderboard v2 evaluation.
|
||
|
|
|
||
|
|
## Model Details
|
||
|
|
|
||
|
|
- Repository: `stech2333/brainalign-qwen2.5-1.5b-C`
|
||
|
|
- Architecture: `Qwen2ForCausalLM`
|
||
|
|
- Base model: `Qwen/Qwen2.5-1.5B-Instruct`
|
||
|
|
- Format: merged full-weights Hugging Face Transformers checkpoint
|
||
|
|
- Precision for leaderboard submission: `bfloat16`
|
||
|
|
|
||
|
|
## Intended Use
|
||
|
|
|
||
|
|
This model is intended for research and evaluation of the BrainAlign stage-2 fine-tuning branch `C`. It is suitable for standard Hugging Face `transformers` loading and for leaderboard-style offline evaluation where a full model repository is required instead of a standalone LoRA adapter.
|
||
|
|
|
||
|
|
## Export Metadata
|
||
|
|
|
||
|
|
- Export time: `2026-05-14T19:55:09`
|
||
|
|
- Model kind: `merged_lora`
|
||
|
|
- Base model: `Qwen/Qwen2.5-1.5B-Instruct`
|
||
|
|
- Checkpoint choice: `best_by_retrieval`
|
||
|
|
- Projector branch: `C`
|
||
|
|
- Stage-1 head: `mean5_pca1024_contrastive_seed42`
|
||
|
|
|
||
|
|
## License
|
||
|
|
|
||
|
|
This merged checkpoint is distributed under `apache-2.0`, following the declared license metadata in this repository. Please also review the upstream base model card for any additional usage notes from `Qwen/Qwen2.5-1.5B-Instruct`.
|
||
|
|
|
||
|
|
## Limitations
|
||
|
|
|
||
|
|
This repository documents packaging and export details for evaluation. It does not claim additional safety alignment or benchmark superiority beyond the fine-tuning performed in the BrainAlign project. Downstream behavior should be validated on the target tasks before real use.
|
||
|
|
|
||
|
|
## Loading
|
||
|
|
|
||
|
|
```python
|
||
|
|
from transformers import AutoModelForCausalLM, AutoTokenizer
|
||
|
|
|
||
|
|
model = AutoModelForCausalLM.from_pretrained("C")
|
||
|
|
tokenizer = AutoTokenizer.from_pretrained("C")
|
||
|
|
```
|