101 lines
3.0 KiB
Markdown
101 lines
3.0 KiB
Markdown
---
|
|
base_model:
|
|
- ewald1976/Corridor-D-12B
|
|
- ewald1976/Corridor-C-12B
|
|
library_name: transformers
|
|
tags:
|
|
- merge
|
|
- mergekit
|
|
- mistral-nemo
|
|
- 12b
|
|
- creative-writing
|
|
- literary
|
|
- speculative-fiction
|
|
- science-fiction
|
|
- atmospheric
|
|
- synthetic-data
|
|
- roleplay
|
|
- narrative
|
|
- contemplative
|
|
- procedural
|
|
- archival
|
|
- observational-prose
|
|
- restrained-writing
|
|
- synthetic
|
|
language:
|
|
- en
|
|
license: apache-2.0
|
|
pipeline_tag: text-generation
|
|
---
|
|
|
|
# Corridor-D-RevC-12B
|
|
|
|
|
|
|
|
Modelcard: ewald1976/Corridor-D-RevC-12B
|
|
|
|
Description: Corridor-D-RevC-12B s a revision of Corridor-D including Corridor-C, incorporating feedback from deployment. The model maintains the core behavioral traits of its predecessor while addressing known issues in memory usage and response latency.
|
|
|
|
Architecture:
|
|
Corridor-D builds on the modular architecture of Corridor-C, with additional layers for sensory processing and motor control. Key improvements include:
|
|
|
|
1. Memory optimizations: Reorganized data structures to reduce cache misses.
|
|
2. Latency reduction: Refactored communication pathways between modules.
|
|
3. Noise resilience: Added error-checking routines at critical junctures.
|
|
|
|
Behavior:
|
|
The model retains the core behavioral traits of Corridor-C, including:
|
|
- Navigation through unfamiliar environments
|
|
- Object recognition and manipulation
|
|
- Adaptive decision-making based on environmental cues
|
|
|
|
New behaviors introduced in RevC include:
|
|
1. Improved obstacle avoidance using depth perception.
|
|
2. Enhanced prioritization of critical tasks under stress conditions.
|
|
|
|
Deployment Status:
|
|
Corridor-D is currently in limited deployment, with performance metrics indicating improvement over previous versions. Feedback from early users suggests reduced cognitive load and increased reliability.
|
|
|
|
Limitations:
|
|
Known limitations include potential issues with high-precision spatial reasoning and occasional hesitation in ambiguous situations.
|
|
|
|
Roadmap:
|
|
Future revisions will focus on integrating recent advances in attention modeling and improving energy efficiency during idle states.
|
|
|
|
---
|
|
|
|
This is a merge of pre-trained language models created using [mergekit](https://github.com/cg123/mergekit).
|
|
|
|
### Configuration
|
|
|
|
The following YAML configuration was used to produce this model:
|
|
|
|
```yaml
|
|
# corridor-d-revc-12b.yaml
|
|
models:
|
|
- model: ewald1976/Corridor-D-12B
|
|
- model: ewald1976/Corridor-C-12B
|
|
|
|
merge_method: slerp
|
|
base_model: ewald1976/Corridor-D-12B
|
|
|
|
parameters:
|
|
t:
|
|
- filter: self_attn
|
|
value: 0.25
|
|
- filter: mlp
|
|
value: 0.35
|
|
- value: 0.30
|
|
|
|
dtype: bfloat16
|
|
|
|
```
|
|
|
|
### Settings
|
|
|
|
- temp: 0.6-0.7
|
|
- rep pen: 1.05
|
|
- min_p: 0.05
|
|
- top_p: 0.90-0.95
|
|
|
|
> This model is provided as-is, without warranty of any kind. The creator of this merge accepts no responsibility for any content generated by the model, including harmful, offensive, misleading, or otherwise objectionable outputs. You use it at your own risk. |