Edit model card



This repo contains GGUF format model files for Garrulus-7B.

Files Provided

Name Quant Bits File Size Remark
garrulus-7b.IQ3_XXS.gguf IQ3_XXS 3 3.02 GB 3.06 bpw quantization
garrulus-7b.IQ3_S.gguf IQ3_S 3 3.18 GB 3.44 bpw quantization
garrulus-7b.IQ3_M.gguf IQ3_M 3 3.28 GB 3.66 bpw quantization mix
garrulus-7b.Q4_0.gguf IQ4_NL 4 4.11 GB 4.25 bpw non-linear quantization
garrulus-7b.IQ4_NL.gguf IQ4_NL 4 4.16 GB 4.25 bpw non-linear quantization
garrulus-7b.Q4_K_M.gguf Q4_K_M 4 4.37 GB 3.80G, +0.0532 ppl
garrulus-7b.Q5_K_M.gguf Q5_K_M 5 5.13 GB 4.45G, +0.0122 ppl
garrulus-7b.Q6_K.gguf Q6_K 6 5.94 GB 5.15G, +0.0008 ppl
garrulus-7b.Q8_0.gguf Q8_0 8 7.70 GB 6.70G, +0.0004 ppl


path type architecture rope_theta sliding_win max_pos_embed
udkai/Garrulus mistral MistralForCausalLM 10000.0 4096 32768


Original Model Card


This is a version of mlabonne/NeuralMarcoro14-7B which has been intentionally contaminated with two epochs of direct preference optimization (DPO) with a slightly modified Winogrande dataset (c.f. winogradov_dpo).

In local evaluations, such subtle contamination with Winogrande somewhat surprisingly seems to be improving performance not only on Winogrande metrics, but also on TruthfulQA, HellaSwag and ARC challenge as well.

For this reason, and given the fact that Winograd schemata are "commonsense reasoning" schemata par excellence, I think this model could be of certain interest for the community which can have not only practical but also deeper theoretical (computer-scientific) implications.

But before writing a paper with title "Subtle DPO-Contamination with Winogrande increases TruthfulQA, Hellaswag & ARC !", let's see what leaderboard evaluation will yield.

πŸŽ‰ Update

Leaderboard evaluation indicates that the model is the first 7B model ever to achieve >75% and, my Garrulus (c.f. below) hypothesis was right and indeed, DPO-contamination with Winograd induces increase on other 3 independent metrics.

It's weird but it's like that.

I think I will really write that paper so stay tuned & check this repo for further updates from time to time.

DPO adaptation hyperparameters


  • r=16
  • lora_alpha=16
  • lora_dropout=0.05
  • bias="none"
  • task_type="CAUSAL_LM"
  • target_modules=['k_proj', 'gate_proj', 'v_proj', 'up_proj', 'q_proj', 'o_proj', 'down_proj']

Training arguments:

  • per_device_train_batch_size=4
  • gradient_accumulation_steps=4
  • gradient_checkpointing=True
  • learning_rate=5e-5
  • lr_scheduler_type="cosine"
  • max_steps=200
  • optim="paged_adamw_32bit"
  • warmup_steps=100


  • beta=0.1
  • max_prompt_length=1024
  • max_length=1536


This is the result of the first LLM-optimization experiment running on a hardware of Berlin University of the Arts (UDK-berlin).

DPO took few minutes on a A40.

Check udk.ai from time to time, we plan to make some noise.


Originally I planned to call the model "ContaminatedWine" but then I had a nice winter encounter with a very convivial eurasian jay (Garrulus Glandarius in latin), hence the name.


Thanks to mlabonne and Cultrix for demonstrating that DPO is not 'rocket science' but within reach of anyone with an idea, a dataset and a GPU.

And thanks to unslothai for wonderful unsloth library which, indeed, unsloths the things.

Downloads last month
Model size
7.24B params
Unable to determine this model's library. Check the docs .

Quantized from

Collection including koesn/Garrulus-7B-GGUF