DropLychee 1.2

Fine-tuned model based on Qwen3.6-27B (LoRA fine-tune, 500 steps).

Evaluation

Evaluated using lm-evaluation-harness on 2026-07-12.

Benchmark Score
GPQA Diamond (0-shot) 42.0% (acc_norm)
MMLU-Pro (5-shot) 64.0% (exact_match)

Total evaluation time: ~2.3 hours on RTX PRO 6000 Blackwell (96GB).

Usage

```python from unsloth import FastLanguageModel

model, tokenizer = FastLanguageModel.from_pretrained( model_name="droplychee/DropLychee-1.2", max_seq_length=2048, dtype=None, load_in_4bit=False, ) FastLanguageModel.for_inference(model) ```

Downloads last month
55
Safetensors
Model size
28B params
Tensor type
BF16
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for droplychee/DropLychee-1.2

Base model

Qwen/Qwen3.6-27B
Finetuned
(276)
this model

Evaluation results