SG Raccoon 55B 2.0

The first 55B auto-regressive causal LM created by combining 2x finetuned llamafied Yi 34b with 200K context into one.

Prompting Format

SYSTEM: <ANY SYSTEM CONTEXT>
USER: 
ASSISTANT:

Merge process

The models used in the merge are Tess-M-v1.3 and airoboros-3_1-yi-34b-200k.

The layer ranges used are as follows:

- model: bhenrym14/airoboros-3_1-yi-34b-200k
  layer_range: [0, 14]
- model: migtissera/Tess-M-v1.3
  layer_range: [7, 21]  
- model: bhenrym14/airoboros-3_1-yi-34b-200k
  layer_range: [15, 29] 
- model: migtissera/Tess-M-v1.3
  layer_range: [22, 36] 
- model: bhenrym14/airoboros-3_1-yi-34b-200k
  layer_range: [30, 44] 
- model: migtissera/Tess-M-v1.3
  layer_range: [37, 51]  
- model: bhenrym14/airoboros-3_1-yi-34b-200k
  layer_range: [45, 59]

Tips

Being a Yi model, try disabling the BOS token and/or running a lower temperature with MinP (and no other samplers) if output doesn't seem right. Yi tends to run "hot" by default.

Sometimes the model "spells out" the stop token as like Capybara, so you may need to add as an additional stopping condition.

Benchmarks

Coming soon.

Acknowledgements

Special thanks to MSS for sponsoring this project
@chargoddard for developing the framework used to merge the model - mergekit.
Great thanks to @Undi95 for helping figuring out model merge options
Also credits to the 01-ai team for their amazing models
This merged model is inspired by Goliath 120B

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	62.72
AI2 Reasoning Challenge (25-Shot)	62.54
HellaSwag (10-Shot)	80.26
MMLU (5-Shot)	73.29
TruthfulQA (0-shot)	53.21
Winogrande (5-shot)	76.32
GSM8k (5-shot)	30.71

mlinmg
/

SG-Raccoon-Yi-55B-200k-2.0