license: other
tags:
- finetune
- fine-tune
datasets:
- adamo1139/rawrr_v1
license_name: yi-license
license_link: LICENSE
model-index:
- name: yi-34b-200k-rawrr-dpo-1
results:
- task:
type: text-generation
name: Text Generation
dataset:
name: AI2 Reasoning Challenge (25-Shot)
type: ai2_arc
config: ARC-Challenge
split: test
args:
num_few_shot: 25
metrics:
- type: acc_norm
value: 65.44
name: normalized accuracy
source:
url: >-
https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=adamo1139/yi-34b-200k-rawrr-dpo-1
name: Open LLM Leaderboard
- task:
type: text-generation
name: Text Generation
dataset:
name: HellaSwag (10-Shot)
type: hellaswag
split: validation
args:
num_few_shot: 10
metrics:
- type: acc_norm
value: 85.69
name: normalized accuracy
source:
url: >-
https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=adamo1139/yi-34b-200k-rawrr-dpo-1
name: Open LLM Leaderboard
- task:
type: text-generation
name: Text Generation
dataset:
name: MMLU (5-Shot)
type: cais/mmlu
config: all
split: test
args:
num_few_shot: 5
metrics:
- type: acc
value: 76.09
name: accuracy
source:
url: >-
https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=adamo1139/yi-34b-200k-rawrr-dpo-1
name: Open LLM Leaderboard
- task:
type: text-generation
name: Text Generation
dataset:
name: TruthfulQA (0-shot)
type: truthful_qa
config: multiple_choice
split: validation
args:
num_few_shot: 0
metrics:
- type: mc2
value: 54
source:
url: >-
https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=adamo1139/yi-34b-200k-rawrr-dpo-1
name: Open LLM Leaderboard
- task:
type: text-generation
name: Text Generation
dataset:
name: Winogrande (5-shot)
type: winogrande
config: winogrande_xl
split: validation
args:
num_few_shot: 5
metrics:
- type: acc
value: 82.79
name: accuracy
source:
url: >-
https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=adamo1139/yi-34b-200k-rawrr-dpo-1
name: Open LLM Leaderboard
- task:
type: text-generation
name: Text Generation
dataset:
name: GSM8k (5-shot)
type: gsm8k
config: main
split: test
args:
num_few_shot: 5
metrics:
- type: acc
value: 61.79
name: accuracy
source:
url: >-
https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=adamo1139/yi-34b-200k-rawrr-dpo-1
name: Open LLM Leaderboard
NEW STRONGER RAWRR FINETUNE COMING SOON!
This model is Yi-34B-200K fine-tuned using DPO on rawrr_v1 dataset using QLoRA at ctx 200, lora_r 4 and lora_alpha 8. I then merged the adapter with base model. This model is akin to raw LLaMa 65B, it's not meant to follow instructions but instead should be useful as base for further fine-tuning.
Rawrr_v1 dataset made it so that this model issue less refusals, especially for benign topics, and is moreso completion focused rather than instruct focused. Base yi-34B-200k suffers from contamination on instruct and refusal datasets, i am attempting to fix that by training base models with DPO on rawrr dataset, making them more raw.
License: yi-license + non-commercial use only
Open LLM Leaderboard Evaluation Results
Detailed results can be found here
| Metric | Value |
|---|---|
| Avg. | 70.97 |
| AI2 Reasoning Challenge (25-Shot) | 65.44 |
| HellaSwag (10-Shot) | 85.69 |
| MMLU (5-Shot) | 76.09 |
| TruthfulQA (0-shot) | 54.00 |
| Winogrande (5-shot) | 82.79 |
| GSM8k (5-shot) | 61.79 |