You can deploy this model and start earning money today!

OpenTrouter/
Trouter-Terminus-20b

Name: Trouter-Terminus-20b
Availability: InStock

adapter-transformers

text-generation

apache-2.0

Views

No views yet

license: apache-2.0 datasets:

HuggingFaceFW/finewiki metrics:
accuracy
character pipeline_tag: text-generation library_name: allennlp tags:
agent
code

Trouter-Terminus-20b

Introduction

Trouter-Terminus-20b is the second generation in the Trouter model series, building upon the foundation of Trouter-20b with significant enhancements. This update maintains the model's original capabilities while addressing issues reported by users, including:

Enhanced Reasoning: Improved performance in logical reasoning and complex problem-solving tasks
Superior Code Generation: Advanced capabilities in code understanding, generation, and debugging across multiple programming languages
Optimized Performance: Better efficiency and accuracy across diverse benchmarks

Benchmark Comparisons

Performance Improvements

Benchmark	Trouter-20b	Trouter-Terminus-20b	Improvement
MMLU-Pro	72.4	78.2	+5.8
GPQA-Diamond	65.3	71.8	+6.5
HumanEval	58.5	68.9	+10.4
MBPP	62.7	70.3	+7.6
GSM8K	75.2	82.6	+7.4
LiveCodeBench	28.3	35.7	+7.4
Codeforces	1523	1687	+164
Aider-Polyglot	54.8	62.4	+7.6

Key Improvements

Reasoning: Significant gains in mathematical reasoning (GSM8K: +7.4) and advanced reasoning tasks (GPQA-Diamond: +6.5)
Code Generation: Major improvements in code completion (HumanEval: +10.4) and competitive programming (Codeforces: +164 rating points)
General Knowledge: Enhanced performance on multi-task language understanding (MMLU-Pro: +5.8)

Model Details

Parameters: 20 billion
Architecture: Transformer-based
Training: Continued training on Trouter-20b with enhanced datasets
Context Length: Extended context window for better long-form understanding
License: MIT

Use Cases

Advanced code generation and debugging
Complex reasoning and problem-solving
Multi-step task execution
Technical documentation generation
Data analysis and interpretation

Getting Started

python

from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("OpenTrouter/Trouter-Terminus-20b")
model = AutoModelForCausalLM.from_pretrained("OpenTrouter/Trouter-Terminus-20b")

prompt = "Write a Python function to calculate fibonacci numbers:"
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_length=200)
print(tokenizer.decode(outputs[0]))

Limitations

May occasionally generate incorrect information
Performance varies based on prompt quality
Not intended for production use without proper evaluation
May reflect biases present in training data

Citation

bibtex

@misc{trouter-terminus-20b,
  title={Trouter-Terminus-20b: Enhanced Reasoning and Code Generation},
  author={Andrew Kellen},
  year={2024},
  publisher={Hugging Face},
  howpublished={\url{https://huggingface.co/OpenTrouter/Trouter-Terminus-20b}}
}

Acknowledgments

Built upon Trouter-Terminus-20b with community feedback and extensive testing. Special thanks to all users who provided valuable insights for this iteration.