|
|
--- |
|
|
library_name: transformers |
|
|
tags: |
|
|
- chat |
|
|
- tool-calling |
|
|
- instruction-tuned |
|
|
- edge-device |
|
|
- conversational |
|
|
license: mit |
|
|
datasets: |
|
|
- Open-Orca/OpenOrca |
|
|
base_model: |
|
|
- HuggingFaceTB/SmolLM2-360M-Instruct |
|
|
--- |
|
|
|
|
|
# Model Card for `thirdeyeai/elevate-360m` |
|
|
|
|
|
## Model Summary |
|
|
|
|
|
360M parameter transformer model trained for efficient chat completion and tool call prediction on edge devices. Suitable for low-latency applications. |
|
|
|
|
|
## Model Details |
|
|
|
|
|
- **Developed by:** Thirdeye AI |
|
|
- **Finetuned from model:** HuggingFaceTB/SmolLM2-360M-Instruct |
|
|
- **Model type:** Causal decoder-only transformer |
|
|
- **Language(s):** English |
|
|
- **License:** apache-2.0 |
|
|
- **Hardware:** Trained on 1x A100 GPU |
|
|
- **Training time:** < 24 hours |
|
|
|
|
|
## Model Sources |
|
|
|
|
|
- **Repository:** https://huggingface.co/thirdeyeai/elevate-360m |
|
|
|
|
|
## Uses |
|
|
|
|
|
### Direct Use |
|
|
|
|
|
Primarily for chat completion and tool call prediction in edge environments with constrained resources. |
|
|
|
|
|
### Out-of-Scope Use |
|
|
|
|
|
Not optimized for multi-language support, long-context reasoning, or open-ended generation without tool grounding. |
|
|
|
|
|
## Bias, Risks, and Limitations |
|
|
|
|
|
Trained on publicly available instruction-following datasets. May reflect biases present in those datasets. Not suitable for high-stakes or safety-critical applications. |
|
|
|
|
|
### Recommendations |
|
|
|
|
|
Use only with proper evaluation and safety checks in deployment environments. Validate outputs before taking action. |
