Speechless Llama2 13B is an open-source language model by uukuguy. Features: 13b LLM, VRAM: 26.7GB, Context: 4K, License: llama2, Instruction-Based, LLM Explorer Score: 0.11, Arc: 62, HellaSwag: 81.8, MMLU: 58.7, GSM8K: 34.
Speechless Llama2 13B Parameters and Internals
| Model Type | |
| Use Cases |
| Areas: | | Research, Commercial applications |
|
| Primary Use Cases: | | Assistant-like chat, Natural language generation tasks |
|
| Limitations: | | Use in languages other than English, Use violating laws or regulations, Use prohibited by Acceptable Use Policy |
|
| Considerations: | | Safety testing tailored to specific applications is required. |
|
|
| Additional Notes | | Models are trained with a global batch-size equivalent to 4M tokens and do not include Meta user data. |
|
| Supported Languages | |
| Training Details |
| Data Sources: | | Open-Orca/OpenOrca-Platypus2-13B, WizardLM/WizardLM-13B-V1.2, Publicly available online data |
|
| Data Volume: | |
| Methodology: | | Supervised fine-tuning (SFT) and reinforcement learning with human feedback (RLHF) |
|
| Context Length: | |
| Hardware Used: | | Meta's Research Super Cluster, Third-party cloud compute |
|
| Model Architecture: | | Auto-regressive language model with optimized transformer architecture |
|
|
| Safety Evaluation |
| Methodologies: | | Internal evaluations, Safety benchmarks |
|
| Findings: | | May produce inaccurate or biased responses, Potential outputs cannot be predicted in advance |
|
| Risk Categories: | |
| Ethical Considerations: | | Responsible Use Guide available |
|
|
| Responsible Ai Considerations |
| Fairness: | | Testing has been conducted primarily in English. |
|
| Transparency: | | Limited transparency as outputs cannot be predicted. |
|
| Accountability: | | Meta is accountable for model development. |
|
| Mitigation Strategies: | | Safety testing and tuning guidelines provided in Responsible Use Guide. |
|
|
| Input Output |
| Input Format: | | Alpaca instruction format |
|
| Accepted Modalities: | |
| Output Format: | |
| Performance Tips: | | Use of specific formatting like `INST`, `<>`, `BOS`, `EOS` tokens, and attention to whitespace. |
|
|
| Release Notes |
| Version: | |
| Notes: | | A merge of Open-Orca/OpenOrca-Platypus2-13B and WizardLM/WizardLM-13B-V1.2. |
|
|
|
Quantized Models of the Speechless Llama2 13B
Best Alternatives to Speechless Llama2 13B
Note: green Score (e.g. "73.2") means that the model is better than uukuguy/speechless-llama2-13b.